ASTROBIOLOGY
Volume 20, Number 10, 2020
Mary Ann Liebert, Inc.
DOI: 10.1089/ast.2020.2241
Subsurface In Situ Detection of Microbes and Diverse
Organic Matter Hotspots in the Greenland Ice Sheet
Michael J. Malaska,1 Rohit Bhartia,2 Kenneth S. Manatt,1 John C. Priscu,3
William J. Abbey,1 Boleslaw Mellerowicz,4 Joseph Palmowski,4 Gale L. Paulsen,4
Kris Zacny,4 Evan J. Eshelman,5 and Juliana D’Andrilli6
Abstract
We used a deep-ultraviolet fluorescence mapping spectrometer, coupled to a drill system, to scan from the
surface to 105 m depth into the Greenland ice sheet. The scan included firn and glacial ice and demonstrated that
the instrument is able to determine small (mm) and large (cm) scale regions of organic matter concentration and
discriminate spectral types of organic matter at high resolution. Both a linear point cloud scanning mode and a
raster mapping mode were used to detect and localize microbial and organic matter ‘‘hotspots’’ embedded in the
ice. Our instrument revealed diverse spectral signatures. Most hotspots were <20 mm in diameter, clearly isolated
from other hotspots, and distributed stochastically; there was no evidence of layering in the ice at the fine scales
examined (100mm per pixel). The spectral signatures were consistent with organic matter fluorescence from
microbes, lignins, fused-ring aromatic molecules, including polycyclic aromatic hydrocarbons, and biologically
derived materials such as fulvic acids. In situ detection of organic matter hotspots in ice prevents loss of spatial
information and signal dilution when compared with traditional bulk analysis of ice core meltwaters. Our
methodology could be useful for detecting microbial and organic hotspots in terrestrial icy environments and on
future missions to the Ocean Worlds of our Solar System. Key Words: Deep-UV spectroscopy—Fluorescence
mapping—Organic detection—Europa—Enceladus—Titan. Astrobiology 20, 1185–1211.
1. Introduction báñez et al., 2018). Similarly, glacial ice from the Greenland
Ice Sheet Project 2 (GISP2) borehole had cellular concen-
Ice is an important environment for life on Earth and trations of 10
4–106 cells/cm3 from 1.5 to 2.5 km below the
possibly elsewhere in the Solar System (Priscu and Hand, surface (Miteva et al., 2009). Priscu and Christner (2004) and
2012; Garcia-Lopez and Cid, 2017). Studies of terrestrial Priscu et al. (2009) used estimates of glacier ice cellular
glacial ice have revealed microenvironments containing vi- concentrations and ice volume to show that the Antarctica
able microbes that are both preserved and actively metabo- and Greenland ice sheets combined represent nearly
lizing (Skidmore et al., 2000; Campen et al., 2003; Miteva 4.4 · 1024 prokaryotic cells, which equates to 4.8 · 10-5 Pg of
and Brenchley, 2005; Miteva, 2008; Liu et al., 2016, 2018). organic carbon—nearly the equivalent carbon biomass of all
These microhabitats exist in a wide variety of icy environ- the liquid freshwater lakes on Earth (Whitman et al., 1998;
ments: snow, glacial ice, deep glacial ice, frozen ponds, sea Priscu and Christner, 2004). Much of the habitable space
ice, etc. (Margesin and Miteva, 2011; Margesin and Collins, inside ice sheets is found in liquid channels and junctions
2019). For example, glacial ice cores collected from the West between ice grains; Barletta et al. (2012) estimated that in-
Antarctic Ice Sheet (WAIS) Divide borehole in Antarctica tergrain liquid water channels in the Antarctic and Greenland
contained prokaryotic cell concentrations from 104 to 106 ice sheet represent 16.7 and 576 km3, respectively, of hab-
cells/cm3 at a depth of 1.7–2.7 km below the surface (Santi- itable space that can support microbial populations.
1Jet Propulsion Laboratory/California Institute of Technology, Pasadena, California, USA.
2Photon Systems, Inc., Covina, California, USA.
3Department of Land Resources & Environmental Sciences, Montana State University, Bozeman, Montana, USA.
4Honeybee Robotics, Altadena, California, USA.
5Impossible Sensing, St. Louis, Missouri, USA.
6Louisiana Universities Marine Consortium, Chauvin, Louisiana, USA.
 Michael J. Malaska et al., 2020; Published by Mary Ann Liebert, Inc. This Open Access article is distributed under the terms of the
Creative Commons License (http://creativecommons.org/licenses/by/4.0), which permits unrestricted use, distribution, and reproduction in
any medium, provided the original work is properly credited.
1185
1186 MALASKA ET AL.
With temperature and pressure conditions similar to the deep et al., 2012; Santibáñez et al., 2019). Barletta et al. (2012)
interiors of ice sheets on Earth, the deep convecting icy crusts of used micro-Raman spectroscopy to show that sulfate and
the Ocean Worlds of the Solar System (Europa, Enceladus, Ti- nitrate concentrations in the channels of glacial ice from the
tan) may also be habitable environments. The deep ice of the GISP2 and Antarctic alpine glaciers were between 104 and
Outer Solar system satellites would be the first potentially hab- 105 times greater than the concentrations measured in the
itable environment encountered by future missions targeting the bulk meltwater.
deep subsurface oceans of those worlds (Marion et al., 2003; For insoluble materials, laboratory experiments show that
Priscu and Hand, 2012; Vance et al., 2018). Exploring the var- the size of solid grains plays an important role, with grains
ious icy habitats in terrestrial subsurface ice will lead to a better <5 mm becoming partitioned into the brine channels, and
understanding of the potential for life in the icy environments on grains >5 mm becoming embedded in the growing ice crystal
Earth and also constrain our understanding of how microbes and (Mader et al., 2006). However, recent work by Santibáñez
organic biomolecules can be preserved in the icy regions of et al. (2019) demonstrated that microbes, both alive and
planetary bodies (e.g., Mars’ poles, Europa, Enceladus, Titan, dead, along with more simple organic materials, will be
etc.) (Priscu et al., 1998, 1999; Price, 2006; Priscu and Hand, incorporated into lake ice during freezing, suggesting that
2012; Boetius et al., 2015; D’Andrilli et al., 2017; Garcia-Lopez more complex partitioning processes may occur. Work by
and Cid, 2017). In particular, understanding how organic ma- Pinto et al. (2019) showed that local concentrations of or-
terials (nutrients, building blocks) and microbial cells in terres- ganic molecules can influence the types of microbes present
trial ice sheets are distributed and interrelated in ice in icy environments such as snow where organic emplace-
microenvironments (Priscu et al., 2007; Margesin and Collins, ment determines the makeup of the microbial community.
2019) will aid in the exploration of these alien worlds. Thus, understanding organic partitioning in ice can provide
The Greenland ice sheet consists of meteorically derived key information about the ice microenvironment that de-
ice that can serve as an analogue for other icy locations in the termines the microbial population.
Solar System. The high altitude (3211 m) and low tempera- Each of these observations and experiments show that ice
tures (-48C average winter minimum to -11C average formation and subsequent modification have the ability to
summer maximum) at the top of the ice sheet represent a partition soluble and insoluble materials and microbes. For
textbook example of dry (minimal liquid) deposition, com- example, partitioning coefficients between ice and water
pression, and modification of grains to form ice (see Lomo- during the freezing process have been used to predict the
naco et al., 2011; Baker, 2019). At the surface, the ice is geochemistry and bacterial density in Vostok Subglacial
porous and consists of a series of interlocked grains known as Lake, which lies beneath *4 km of ice in East Antarctica
firn, whereas deeper, the grains become altered as the pore (Priscu et al., 1999). However, what is not known is the
spaces close off and firn converts to nonporous glacial ice. ultimate partitioning and distribution of microbes and or-
Previous work by Lomonaco et al. (2011) shows that air ganic materials in glacial ice (i.e. in situ) at fine spatial
pockets seal off completely from exchange with the surface at scales in microhabitats (Margesin and Collins, 2019). It is
a depth of 69 m (referred to as the lock-in zone) whereas the likely that the first samples for the ocean worlds in the outer
pore close-off occurs at 81 m. These same processes can Solar System will be obtained from their icy shells. Hence,
occur on Mars where deposited ice material accumulates and it is imperative that methods be developed and deployed to
compresses. They may also occur on Saturn’s moon En- these frozen worlds that can characterize organic matter
celadus (and possibly Jupiter’s moon Europa) in areas near within the ice, which then can be used together with a
jets where ballistically ejected water ice accumulates back knowledge of crustal dynamics to predict conditions in the
onto the surface. It is possible that the entire history of ejected underlying deep subsurface ocean.
organic and other materials of Enceladus’ jets could be re- Key questions we attempted to answer in our present
corded in these near-jet re-deposition terrains, just like the study include: Are the naturally fluorescent microbes and
snowfall record on Earth. Although the lower gravity and other organics within the ice confined to discrete layers due
lower temperatures compared with Earth will affect the cor- to initial deposition? Are higher concentrations also associ-
responding depth where porous firn transitions to nonporous ated with bubbles or emplaced in other discrete spots? Does
ice, the same fundamental physical and chemical principles the transformation from firn to glacial ice alter the distri-
will apply (see Vance et al., 2018 for predicted T and P bution of fluorescent materials? Is the distribution of fluo-
curves for the Ocean Worlds in the outer Solar System). rescent materials the same in firn ice as it is in glacial ice?
Deposition and compression, but minimal lateral movement In the past, these questions were difficult to answer.
and fluid percolation, make the top of the Greenland ice sheet Traditionally, chemical and microbial analysis of ice cores
an ideal location to study the processes involved in the is accomplished by extracting the ice core, then melting the
transition of snow to firn to glacial ice, and the materials ice core by continuous melting or melting of core sections,
trapped within (Meese et al., 1994; McGrath et al., 2013). followed by chemical analysis of the melted sample at
Research on natural and laboratory terrestrial ice has varying resolution (Christner et al., 2005; Miteva and
shown that microbes in ice are found localized in brine Brenchley, 2005; Tung et al., 2005; Yung et al., 2007;
channels and triple junctions between ice grains (Price, Miteva et al., 2009; D’Andrilli et al., 2017; Santibáñez
2000; Junge et al., 2001; Mader et al., 2006; Barletta and et al., 2018). Unfortunately, this procedure results in loss of
Roe, 2012; Barletta et al., 2012). For water-soluble mate- high spatial resolution, dilution of analytes (often to unde-
rials such as salts, as water is incorporated into growing ice tectable levels) that are concentrated in microenvironments,
grains, soluble materials are excluded and pushed into a and loss of critical information for discerning the mecha-
network of micron-scale inter-grain channels that surround nisms of emplacement and distribution. Any details at scales
the ice grains (Price, 2000; Barletta and Roe, 2012; Barletta finer than that used to create meltwater that is sufficient for
GLACIAL ICE EXPLORATION 1187
analysis are lost. Such information is critical to understanding classified as ‘‘ice cap,’’ with no month during the year having
the temporal sequence of paleoclimate in meteoric ice and for a mean monthly temperature exceeding 0C, with exceptions
predicting conditions in underlying water bodies such as in 2019 (after the completion of our expedition), 2012, 1889,
subglacial lakes on Earth and the ocean worlds of our Solar and 7 other times between 1250 and 750 AD (Meese et al.,
System. In addition, during core extraction and sample han- 1994; McGrath et al., 2013). During the course of our field-
dling there is the risk of physical damage from melting, work (June 4 to July 10, 2019), the temperature at Summit
cracking from pressure release, and contamination from ex- Station averaged -10.5C (13.1F), with a minimum of
traction, bagging, and transport (Tung et al., 2005). Decon- -26.9C (-16.4F) and a maximum daily temperature of
tamination procedures exist for ice cores, but they require 0.9C (33.6F). Humidity during this same period averaged
removing *2 cm of the outer ice core, resulting in significant 82%, with a minimum of 55% and a maximum of 98%.
sample loss (Christner et al., 2005). For exploration of
planetary surfaces beyond Earth, extraction of ice cores and 2.2. Drill site location
shipment back to a laboratory is complex and likely impos-
The WATSON team elected to begin its drill efforts in a
sible; thus, an in situ measurement technique is desirable.
pre-existing borehole to maximize the time spent interro-
The overarching objective of our work is to determine the
gating deeper glacial ice, as opposed to drilling in a new
fine-scale distribution of cellular and noncellular organic
location through firn. Our team drilled and extended the
matter without ice core extraction, transport, melting, and
‘‘Baker Borehole,’’ a borehole located 6.75 km from Sum-
the associated risks of contamination. To address this ob-
mit Station at N72.59781, W38.62563 (August 2018).
jective, we developed a deep-ultraviolet (DUV) fluorescence
During a project led by Ian Baker (Dartmouth College), the
Raman mapping spectrometer for in situ analysis (Eshelman
borehole was drilled in 2017 with a UWM-IDDO Eclipse-
et al., 2019). Spatially resolved DUV fluorescence spec-
Badger drill. It was ‘‘dry drilled’’—meaning that no drilling
troscopy of ice is a powerful tool for the identification and
fluid was used to prevent lithostatic collapse of the borehole.
localization of different molecular species based on elec-
Baker (2019) acquired firn cores to a depth of *85 m to
tronic transitions (Rohde and Price, 2007; Bhartia et al.,
study firn densification mechanisms and microstructural
2008, 2010; Rohde et al., 2008; Price, 2009; Rohde, 2010;
evolution (Ian Baker, personal communication; see also
Barletta and Roe 2012; Barletta et al., 2012; Eshelman
Lomonaco et al., 2011). At the end of their campaign, the
et al., 2019). For in situ exploration, we coupled a DUV
borehole was capped with a cover to prevent snow accu-
fluorescence mapping spectrometer with a wireline robotic
mulation in the borehole. In June 2019, the WATSON drill
drill system to create a down-borehole instrument-drill
team extended the Baker Borehole from 84.5 to 110.5 m
combination referred to as WATSON (Wireline Analysis
with the Planetary Deep Drill prototype (Zacny et al., 2016;
Tool for the Subsurface Observation of Northern ice sheets).
Mellerowicz et al., 2018). On arrival at the drill site, our
This combination is a departure from traditional ice explo-
initial inspection revealed that the original hole had main-
ration where a sample is acquired and delivered to an in-
tained its integrity and depth—therefore, no borehole
strument: WATSON brings an instrument to an in situ
reaming was required. The diameter of the extended bore-
sample.
hole was 11.18 cm (4.4†). Down-borehole camera images
In July of 2019, we used WATSON to drill a borehole
(not part of WATSON instrument) detected a melt lens at
and make fine-scale in situ measurements of the borehole
3.28 m depth, which we interpreted to be due to the 2012
wall to determine the distribution of fluorescent organic
surface melt (McGrath et al., 2013).
materials in firn and glacial ice to a depth of 107 m near
Summit Station, Greenland. Our investigations at Summit
2.3. WATSON instrument and drill combination
Station provide new information on the processes that affect
localization (and concentration) of organic and putative WATSON consists of a DUV fluorescence Raman in-
biosignatures during meteoric deposition and compaction. In strument coupled to the Honeybee Robotics Planetary Deep
addition, our efforts advanced detection and characterization drill (Fig. 1). The 4 m long, 10 cm in diameter drill system
techniques that could be used in a laboratory setting to combines a coring bit, power head, retractable anchor feet,
identify hotspots in extracted ice cores for further analysis. and retractable skids at the lower end of a drill string, cou-
pled to a DUV fluorescence Raman instrument with a me-
2. Methods chanical stage for the optical mirror, as described in the work
of Eshelman et al. (2019), at the upper end. For deployment
2.1. Summit Station, Greenland
(Fig. 2), WATSON is suspended on a power/communication
Summit Station was established in 1989 as part of the tether from a drill platform and lowered down the borehole
GISP2 drill site. It is located near the apex of the Greenland from the surface. An encoder attached to the tether measures
ice sheet, just inside the southwestern boundary of the North the distance traveled and an onboard inertial measuring unit
East Greenland National Park (72.6N, 38.5W, elevation determines the rotation (thus orientation) inside the borehole
3211 m [10,535¢] EGM96, August 2019) (S. Dorsi, personal relative to magnetic north. The communication tether allows
communication; Summit Guide, 2019). The Station is cur- commanding and data transfer with the surface. The drill
rently maintained by the U.S.-based CH2M HILL Polar system includes onboard electronics, sensors, and computers
Services, with support from the National Science Foundation, to monitor the drilling operation.
for the purpose of providing year-round support for scientific Before initial insertion into the ice borehole, the entire
research into glaciology, meteorology, atmospheric chemis- drill and spectrometer was thoroughly cleaned with 60–70%
try, and a wide range of other fields requiring long-term en- ethanol in water and wiped dry. The drilling operation starts
vironmental observations. The climate at Summit is currently with lowering the drill to the bottom of the borehole. Once
FIG. 1 (A–C) Views of the instrument drill combination. (A) Schematic drawing of the Honeybee Planetary Deep Drill
and WATSON DUV mapping spectrometer. Mannequin for scale. (B) Image of actual constructed drill and stand. (C)
Image of WATSON DUV mapping spectrometer. Left side of image shows uncovered components, and right image shows
instrument in down-borehole configuration with the protective tube in place. The optical window can be seen at center right.
DUV, deep-ultraviolet; WATSON, Wireline Analysis Tool for the Subsurface Observation of Northern ice sheets. Color
images are available online.
FIG. 2. Image and schematic of operations showing in-borehole operation and data flow. (A) Image of WATSON drill
instrument in drill tent at Summit Station, Greenland. The upright silver tube at the center is the WATSON instrument.
Optical window at lower center in image. The rest of the drill is inside the borehole at lower center in image. (B) Cutaway
schematic showing drill-instrument combination in the ice borehole. The WATSON instrument is indicated by a blue
rectangle in the schematic. The laser illumination (not to scale) is indicated as a purple beam penetrating into ice and hitting
a target. During a point cloud measurement, the instrument is moved vertically as the laser is illuminated along the length of
the borehole. During a map, the instrument is stationary, and the laser beam is moved by a series of internal windows to
build up a raster map product. (C) Graphic of a map product showing a portion of the borehole. Green squares indicate
pixels of interest. (D) Cartoon simulating a spectral response extracted from one of the map pixels. Color images are
available online.
1188
GLACIAL ICE EXPLORATION 1189
at the bottom, the coring bit cuts a 50 cm-long ice core in There were five gaps in depth coverage: from 0 to 3.3, 8.96 to
approximately 1–2 minutes. The drill is then lifted off the 9.75, 10.48 to 19.81, 88.16 to 88.56, and 99.66 m to 99.78 m.
hole bottom and during this step, core dogs cut and liberate As stated earlier, point clouds interleaved dark and active
the ice core from the substrate. The drill is subsequently spectra. Since the entire drill-instrument combination, but not
lifted out of the hole where the core is removed, labeled, and instrument mechanical stage, was in motion, the laser fired
stored for further analysis. The up and down operation is through the same section of the window during the scan.
performed sequentially until the required depth is reached. However, there was a wavelength-dependent background
During our drilling campaign, the average drilling rate was signal that needed to be removed from the acquired instru-
6.5 m/day. We collected a total of 26 m of ice core, while ment signal intensity. This background also decreased during
exposing 26 m of fresh borehole wall for the WATSON the course of a particular cloud or line scan run. During post-
instrument to interrogate. The average density of ice cores processing, we created a synthetic background for each point
collected for this project, from 84.5 to 110.5 m, was 0.84 g/ by determining average spectra of 10 successive laser shot
cm3; this is consistent with densities of glacial ice reported reads after *500 laser shots and at the end of the run; the
in the work of Bender et al. (1997), which have density standard deviation of the 10 shots at each location was also
values of 0.81–0.84 g/cm3 and air volumes of only 10–15%. recorded. After 500 shots, the background drift was in a linear
During instrument operation, a pulsed (40ms) DUV (ex- regime. For each point, a synthetic background was created
citation wavelength: 248.6 nm) hollow-cathode NeCu laser based on a linear interpolation of the average values at the
was fired at 40 Hz through a series of optics that cleaned and two regions in the point cloud data, and this background was
shaped the laser beam and focused it through a fused silica subtracted from the non-bias corrected linear spectra for each
optical window onto the borehole wall. This DUV illumi- laser shot. (Bias offsets are already present in the uncorrected
nation causes aromatic-containing organics to fluoresce at background.) The background generally had increased signals
electronic transitions diagnostic of the electronic states of at both 324.6 and 341.2 nm and was never larger than 2400
the particular molecule or mixture of molecules (Bhartia counts in any of the recorded times of any of the point clouds.
et al., 2008; Rohde, 2010; Eshelman et al., 2019). The Supplementary Data S1 contains the full background-
fluorescence emission signal is collected by the same illu- corrected point cloud dataset.
minating objective lens and reflected into a spectrometer
coupled to a photomultiplier tube (PMT) array gated to the
2.5. Map data acquisition and processing
duration of the laser pulse. The spectral range observed is
275–450 nm with 32 equal bins. The instrument has an in- Maps were acquired by locking WATSON at a constant
ternal stage that allows raster scanning through an optical depth inside the borehole by using metal skids and a ser-
window. During scanning, operations involved collecting a pentine raster line scan of the optical path across a target
dark (no laser) and an active acquisition during the 40ms area. The scanning motor and laser repetition rate were
laser shot. This allowed subtraction of a dark background at synched such that the maps were prepared with a spatial
a time close to the laser shot. Data collected included fluo- scale of 100 mm per pixel. Typical map dimensions were
rescence intensity from each wavelength bin, depth, rota- 1 cm horizontal · 4 cm vertical and 3 cm horizontal · 4 cm
tion, and a timestamp, as well as a flag for whether the signal vertical. As for the point cloud, raster maps signal acquisi-
was collected during a laser shot or as a dark. During point tion interleaved dark and active spectra. After data collec-
cloud operation, we favored slowly lowering the drill with tion, the laser-illuminated instrument counts were subtracted
skids in place to ensure rotational stability; however, there from the dark bias instrument counts and a multiband raster
were times we acquired point clouds at a lower resolution map was created.
where we allowed the drill to freely rotate inside the bore- During map analysis, the background spectra was re-
hole. During detailed raster scanning, skids were engaged moved for each identified spatial region of interest (ROI) in
for rotational stability, and tether playout was prevented by the mapped region. To do this, we selected two multi-pixel
using a brake to maintain a constant depth (Fig. 2). blank areas for each feature. These were approximately
equidistant and were above and below the feature, usually
within 5 pixels (0.5 mm) distance. The blank pixel values for
2.4. Point cloud data acquisition
each area were averaged, and the two sets were interpolated
Point clouds were acquired as a series of successive laser (averaged since equidistant to the target feature) to create a
shots as the drill-instrument system was lowered or raised at synthetic predicted background at the location of the iden-
a constant rate with no movement of the internal mechanical tified feature. The synthetic predicted background was
stage. The physical spacing of shots varied between point subtracted from the instrument response to determine the
clouds acquisitions as the laser firing rate and drill move- actual spectral response. The maximum standard deviation
ment rate allowed varying spacing depending on desired between the two blank areas was set as the noise level.
resolution. We used sparse point shots with a post spacing During ROI determination, only those pixels above 3· noise
varying between 5 mm for rapid surveys and 100mm for level for at least one of the bands were used to define the
more detailed spatial mapping. The points were thus dis- feature outline. For detailed spectral characterization, only
tributed along a nearly vertical linear track during drill de- those pixels that were 10 · noise level for the lambda max
scent or ascent. During signal acquisition, the PMT voltage (lmax) and the main background peaks (either 324.6 or
potential was kept constant at 800 mV. 341.2 nm) were selected for display. Supplementary Data S5
We acquired 32 point clouds in the borehole covering the contains the complete set of points from the ROIs.
length of the borehole from 3.3 to 105.6 m depth. We targeted Later spatial calibration using a resolution target deter-
borehole depth coverage rather than rotational coverage. mined that acquired maps were compressed in the z-direction
1190 MALASKA ET AL.
(y-direction in the xy maps of Fig. 9). Therefore, the di- values within the bin used. The spectra were then normal-
mensions in this direction were only 0.7· the actual value. ized to the highest response in the WATSON spectral
We compensated for this during angle and aspect ratio range (275–425 nm), and this was used for comparison of
measurements by importing the uncompressed image into WATSON spectra.
Photoshop (Adobe), spatially stretching the image in the
y-direction, and recording the key dimensions and values 3. Results
with Photoshop tools. The map presented in Fig. 9 and the
3.1. Point cloud data analysis
dataset in Supplementary Data S5 are spatially uncorrected
in the depth direction—they represent the actual pixel co- We identified points of interest in the point cloud data by
ordinates directly obtained from the WATSON instrument. searching for a clear discernable pattern when subtracted
from the background spectrum. In general, this meant that at
least three consecutive band responses were at least 10 times
2.6. Fluorescence measurement of laboratory
the maximal standard deviation from the two regions used
comparison standards
for background determination. An example plot for the
For comparison spectra, laboratory-acquired Excitation- 385.3 nm wavelength band of the Point Cloud 17 dataset is
Emission Matrices (EEMs) were obtained by using either shown in Fig. 3. Points above the 10 · noise level (above the
a Hitachi F-4500 spectrometer or a Horiba Jobin-Yvon red line in the figure) were deemed potential points of in-
Fluoromax-4 Spectrofluorometer, equipped with an Xe lamp terest, and signals from neighboring bands were examined
light source and a pathlength of 1 cm. The Hitachi F-4500 had further.
an excitation wavelength scanning from 200 to 400 nm in Table 1 presents the summarized data for all 32 point
5 nm increments and intensity measurements from 200 to clouds, including the hit rate for each 4000 laser shot cloud
600 nm every 2 nm with a scan speed of 1200 nm/min. reported as the number of points (laser shot returns) of in-
Samples scanned by the Hitachi instrument included pure terest per 1000 laser shots. The hit rate value for each point
chemical standards as well as whole bacterial cultures. For cloud varies from a maximum of 73–0. The combined point
pure chemical standards, homogeneous aqueous solutions cloud dataset represents the instrument response from
were made and scanned; whereas for whole bacterial sam- 127,999 laser shots. Of these shots, only 1039 points were
ples, culture suspensions in deionized water were used. determined to be of interest for an overall ratio of targets of
Samples run on the Hitachi instrument included naphthalene, interest to shots fired of 0.8%. Supplementary Data S2
tryptophan, phenanthrene, anthracene, pyrene, methyl-1- contains the spectral data for all the points of interest
pyrene, fluoranthene, perylene, Shewanella oneidensis MRI identified from the point clouds.
whole cells, and Bacillus pumillus whole cells. Spectral data from the 1039 points showed a distinctive
The Horiba Jobin-Yvon Fluoromax-4 instrument used spectral pattern only near 84.9 m. This was very close to the
excitation wavelengths from 240 to 450 nm scanned over depth at which new drilling into the borehole took place. We
10 nm intervals, with emission responses measured from 300 deemed this as a possible impurity and, if the 239 points
to 560 nm recorded in 2 nm increments. Samples run on the from this region are removed from the bulk analysis, we
Horiba instrument included vanillic acid, p-benzoquinone, obtain a corrected bulk hit rate of 0.6%. As can be seen from
Suwannee River and Pony Lake fulvic acid samples, and an examination of Table 1, the actual rate varies from point
‘‘Antarctic bacterium’’ isolate. Samples of the chemical so- cloud to point cloud. Two of the point clouds had zero
lutions were prepared to a 20 mg/L stock solution, shaken for points of interest (out of a 4000 laser shots), whereas one of
24 h, and placed into a quartz cuvette; spectra were collected the point clouds had a hit rate of 2.65% (this point cloud did
at 25C. The ‘‘Antarctic bacterium’’ isolate is an extract from not have any of the possible impurities in it). No obvious
a sample collected from a supraglacial stream on the Cotton correlation was noted between post-spacing and observed
Glacier, Antarctica, and is described in the work of Dieser hit rate. Our results suggest that distribution of fluorescent
et al. (2019). The two fulvic acid reference samples were materials in the ice is not uniform and that, on average,
prepared as 2 mg/L solutions in ultrapure Milli-Q water. The many locations need to be interrogated at the 100 mm scale
Pony Lake fulvic acid reference sample is an Antarctic mi- to detect fluorescent materials.
crobial organic matter end-member, originating from Pony It should be noted that the example in Fig. 3 (from point
Lake, Antarctica, and was obtained from the International cloud 17), taken at a depth range of 93.549–94.161 m and a
Humic Substances Society; the Suwannee River fulvic acid rotation orientation of 48 degrees relative to magnetic N,
standard is a continental organic matter end-member (Oke- had 50 identified features. In contrast, a similar transect
fenokee Swamp), and it was also obtained from the Interna- (point cloud 18) at nearly the same depth range, from 93.549
tional Humic Substances Society. Corresponding UV- to 94.217 m, but at a rotation orientation of 33 degrees
absorbance spectra for these samples were also collected for relative to magnetic N, had only four identified features.
EEMs post-processing analysis in accordance with protocols With a borehole diameter of 11.176 cm and a circumference
in the works of D’Andrilli et al. (2017) and Bhartia et al. of 35.11 cm, a difference of 15 degrees corresponds to a
(2008). distance along the borehole’s curved wall of 1.46 cm. Thus,
For compatibility, we extracted only the data coming even at the same depth, a slightly different track, and small
from the 250 nm excitation wavelength from the three- cm-scale horizontal distance, can provide different results
dimensional fluorometer instruments (closest to the 248 nm (Fig. 12). This observation had previously been noted for
excitation used in the WATSON drill-instrument combina- extracted and laboratory-scanned sections of WAIS divide
tion). The recorded spectral emission responses were binned and GISP2 ice samples by Rohde (2010); here, we note a
to match the WATSON wavelength binning and average similar observation in situ.
GLACIAL ICE EXPLORATION 1191
FIG. 3. Example of background-subtracted point cloud data from WATSON. This plot shows the 385.3 nm data for Point
Cloud 17 and was acquired at a depth from 93.5 to 94.2 m below the surface and with an average post-spacing of 153mm.
Data for photomultiplier tube responses for 4000 laser shots are represented here. For clarity, only data for the 385.3 nm
band are shown. The red line indicates the value for 10· the maximum standard deviation from the regions used for
background determination. At this wavelength band, potential points of interest cluster at 93.70, 93.73, and 93.79 m. Color
images are available online.
Figure 4 shows the distribution of identified points of any materials went to the bottom of the borehole, they would
interest by 10 m depth bins. It should be noted that more be emplaced at this depth. In addition, although we worked
point cloud data (more laser shots) were acquired at deeper in a manner to reduce introduced contaminants, our drilling
locations; the overall number of laser shots in each depth efforts to 110 m depth before scanning may also have in-
bin is indicated in Fig. 4 as purple diamonds. In general, troduced organic matter contamination, particularly at
the number of points of interest per 1000 laser shots is shallower depths that were the most frequently traversed
relatively constant from 0 to 105 m, but with higher fre- during the drilling operation. However, we note that the
quencies in the 80–90 m depth bin, the 30–40 m depth bin, sharp zone containing the Spectral Type L385_d_complex
and from the surface to 10 m depth. From visual inspection at 85 m (see Section 3.3 for Spectral Type naming con-
of the borehole images, the firn-ice transition was in the vention) contaminant suggests that although this material
80–90 m depth range. was indeed emplaced at some point, it remained localized
The reason for the significantly elevated hit rates at 30–40 even after repeated drill ascents and descents. This suggests
and 80–90 m is unclear. We also note an enhancement of hit that our drilling and scanning operations did not affect
rate frequency toward shallower depths. From our down- material emplacement.
borehole imaging, we noted a melt lens at 3.28 m depth that One possible explanation for the variable hit rates by
we interpret as resulting from the surface melting in summer depth bin is that the deposition of material has changed over
of 2012 (McGrath et al., 2013). Our most shallow scan time. The depth bin containing the firn ice boundary at 80–
(Point cloud map 1, Table 1) was just below that level and 90 m corresponds to material deposited approximately 205–
that particular scan did not have a higher than normal hit 245 years ago. The age values are extrapolated and adjusted
rate. However, it also should be noted that the 1889 melt by using data from the work of Mayewski (1999); Yung
period, which corresponds to roughly 58 m depth, did not et al. (2007) found that ice at 94 m corresponds to ice de-
correspond to an increased hit rate in the 50–60 m depth bin posited 265 years ago as measured in 2007. Anthropogenic
(depth correlation adapted from GISP2 data from inputs increased after the industrial revolution; as an ex-
Mayewski, 1999). The borehole was previously drilled in ample, the amount of polycyclic aromatic hydrocarbon
2017 to study the physical structure of the ice, and at that (PAH) molecules detected in Greenland ice cores increases
time, no specific organic matter decontamination proce- significantly due to anthropogenic inputs starting from 1930
dures were taken. The 80–90 m-depth bin contains the 85 m to present (Kawamura et al., 1994; Gabrielli and Vallelonga,
maximum depth attained by the Baker team in 2017; thus, if 2015), which we estimate corresponds from *44 m depth in
1192 MALASKA ET AL.
Table 1. Table of Point Cloud Datasets Collected Sorted by Depth
Point Start rotational Final rotational Average No. of Hit rate
cloud Start Final orientation (degrees orientation (degrees post-spacing identified per thousand
map depth (m) depth (m) from magnetic N) from magnetic N) (lm) features laser shotsa
1 8.964 3.353b 34 355 -1403b 42 10.5
2 9.752 10.482 68 62 183 64 16
3 19.811 20.295 78 69 121 37 9.25
4 20.002 26.297 61 57 1574 11 2.75
5 26.41 47.687 44 66 5321 106 26.5
6 29.144 35.287 130 173 1536 40 10
7 39.819 46.352 154 148 1634 10 2.5
8 48.207 68.368 51 75 5042 10 2.5
9 55.219 61.395 67 91 1544 20 5
10 65.338 72.331 3 186 1749 6 1.5
11 68.913 88.162 66 85 4813 25 6.25 (6)c
12 76.966 86.727 323 350 2441 67 16.75
13 83.75 84.183 35 35 108 0 0
14 84.549 85.048 34 34 125 292 73 (13.5)c
15 88.566 94.043 73 67 1370 36 9
16 89.225 93.879 4 62 1164 19 4.75
17 93.549 94.161 48 48 153 50 12.5
18 93.549 94.217 33 33 167 4 1
19 93.551 94.098 225 224 137 3 0.75
20 93.8 99.402 116 108 1401 9 2.25
21 94.154 94.67 48 48 129 3 0.75
22 94.661 95.135 48 48 119 14 3.5
23 98.05 98.685 46 46 159 83 20.75
24 98.05 98.565 7 10 129 4 1
25 98.556 99.126 20 23 143 7 1.75
26 99.119 99.66 27 29 135 1 0.25
27 99.788 102.916 100 92 782 77 19.25
28 100.001 105.609 11 15 1402 6 1.5
29 104 104.659 47 47 165 4 1
30 104.661 105.289 47 47 157 6 1.5
31 105.28 105.94 323 335 165 7 1.75
32 105.281 105.885 47 47 151 0 0
aEach point cloud consisted of 4000 laser shots.
bPoint cloud 1 was acquired during lifting of the drill. All other point clouds were acquired during descent.
cNumbers in parentheses indicate derived values after exclusion of Spectral Type L385_d_complex features that were only found near
85.5 m and were ascribed to an impurity.
our borehole to the surface (adapted from GISP2 data from unconsolidated boundaries can create an effective interro-
Mayewski, 1999). In particular, fluorescent airborne micro- gation volume that is brightly illuminated nearer the surface;
plastics could be responsible for some of the detected signals this also increases the volume of fluorescence emission that
at shallower depths, and recent studies have shown that mi- can be received (Rohde, 2010). In contrast, glacial ice may
croplastics are present in arctic snow samples (Bergmann have longer pathlengths (light penetration can travel deeper
et al., 2019); plastic production has increased from an annual and straighter into the ‘‘clearer’’ ice) before a fluorophore is
production of 2 MT in 1950 to 380 MT in 2015 (Geyer et al., detected; only fluorophores that lie directly in the beam path
2017), which would correspond to depths in our borehole of will be illuminated (Rohde, 2010; Eshelman et al., 2019).
*38 and 2 m below the surface, respectively. Another factor
is potential degradation of some of the chemical signals with
3.2. Multipoint features
increasing depth; analysis by coupled gas chromatography-
mass spectrometry of snow pit samples at Summit Station by Of the 1039 identified points of interest, 308 (30%) were
Jaffrezo et al. (1994) suggested that certain PAH molecules solitary features composed of a single point where consecu-
degraded with time and depth. Yet another possibility is the tive laser shots did not show a significant signal. The majority
difference in measurable optical properties between glacial of points were associated with one or more consecutive shots
ice and firn. Differences in firn and glacial ice physical and had spectrally related signals. We identified 80 clustered
characteristics such as reflective surfaces and air bubbles features, listed in Supplementary Table S2 by depth from the
may promote or hinder detection of ice features near the surface. Many of these features had only a small number of
borehole surface (e.g., reflective or scattering firn surfaces consecutive points: 38 were composed of 2 consecutive
may make organic matter detection by this method easier points, 18 of 3 consecutive points, and 7 of 4 consecutive
compared with more compressed ice) (Rohde, 2010). For firn points. However, some had a large number of associated
ice, the higher amount of scattering from bubbles and other points—entry 52 was a feature composed of 237 consecutive
GLACIAL ICE EXPLORATION 1193
FIG. 4. Hit rate frequency per 1000 laser shots as a function of depth. Shaded bars show hit rates of all spectral types,
orange colored bars show hit rate if data from the putative impurity, Spectral Type L385_d_complex, is excluded. The
purple diamonds and lower x-axis show the number of laser shots fired in each depth bin. The zone with the firn-glacial ice
transition (80–90 m) is shown by red dashed lines. Color images are available online.
points. We interpreted consecutive spectrally similar laser detailed spectra of points for this entry shown in Supple-
shots as large extended features; the spectral similarity of mentary Fig. S1), which consisted of 13 consecutive spec-
these provided corroborating evidence that the measured trally similar shots with a 5.3 mm spacing. In Fig. 5, we also
signals are not random spikes in the data. However, due to show the data from post-spacings <0.2 mm by using 1 mm bin
the varying laser post-spacing in the point clouds, a larger sizing. Even with this restricted dataset, both distribution
number of consecutive points does not correspond to a patterns in Fig. 5 suggest a skew to smaller-sized features.
larger spanned distance. Supplementary Table S2 lists the The signal intensities at the lmax spanned over four orders
maximum estimated size of the features. of magnitude (Supplementary Data S3 and Supplementary
We calculated the maximal estimated apparent size by Fig. S2). Weak signals (<2000 instrument counts) predom-
assuming that the laser was scanning in a straight line with inate over stronger ones. For example, 70% of all points of
uniform spacing and that the size extended almost one post- interest in the point cloud dataset are within 3000 instrument
spacing to the neighboring post just before and just after the counts of the background, and 95% of all points are below
feature along the line. A potential overestimate of >10.6 mm 20,000 background-subtracted instrument counts. No clear
could occur in cases where the shot post-spacing is correlation between average signal intensity and estimated
5.321 mm—this overestimate would occur in the case where feature size was noted. However, one of the largest features,
the detected laser shot was right at the margin of an ex- entry 77 in Supplementary Table S2, had considerably
tended feature but our estimate extended to the next posting. higher intensity than most other signals. It alone accounted
Given this caveat, the binned distribution of maximally for a large percentage of the overall signals detected above
estimated size of all the features in the point cloud data is 20,000 counts (see signal intensity distribution plot in
shown in Fig. 5. For single-point features, the maximum size Supplementary Fig. S2). The average signal intensities of
was estimated to be 2· the post spacing—it is likely that this the point cloud dataset as well as the raster map dataset
is an overestimate based on data from our higher resolution followed a rough power law, with many features with low
raster map data (see Section 3.4). Due to the variable shot average signal intensity and only a few features with high
spacing mentioned earlier, only some of the higher-resolution signal intensity (Supplementary Fig. S2).
data could be placed in bins smaller than the 10–15 mm bin. The rough distribution of signal intensities has important
Even with this caveat, a clear drop off in larger-sized features implications for future scanning instrumentation or instru-
can be noted in the graph. There are 20 features (out of 388, ment modifications. For the point cloud dataset, if the in-
thus 5%) larger than 15 mm. The largest feature by distance strument sensitivity was decreased so that only peaks above
spanned was 74.5 mm, entry 38 (Supplementary Table S2, the 5000 count level were considered significant, then only
1194 MALASKA ET AL.
FIG. 5. Distribution of estimated size of features identified from point cloud measurements. The size estimate is based on
laser spot spacing and the number of points detected (see Section 3.2). The entire dataset of 388 total features (this includes
both multipoint features and single points of interest) is shown in orange. For a single feature with the largest post-spacing
(5.321 mm for Point Cloud 5), a single point could span 10.642 mm before detection by a second laser shot. Thus, a single
point from this Point Cloud could indicate a feature up to 10.6 mm wide and would go into the 10–15 mm bin. A subset
dataset (purple bars) represents the data that only have higher resolution post-spacing <0.2 mm. In that case, a single pixel
feature would only go into the 0–1 mm bin. Despite the post-spacing variation, both distributions show a skew toward
smaller features. Note the log scale on the y-axis. Color images are available online.
10% of the signals would be detected; the corresponding hit lengths, single-ring aromatic molecules with limited conju-
rate would also drop by an order of magnitude. gating or electronic-modifying functional groups have
fluorescence emission maxima <300 nm. Larger conjugated
or fused aromatic systems with three or more rings have
3.3. Spectral diversity of selected points
fluorescence emission maxima >400 nm. Bicyclic systems,
The selected points from the point cloud data were clas- including the indole ring of the amino acid tryptophan, have
sified into several Spectral Types based primarily on lmax as fluorescence emission maxima between 325 and 375 nm.
well as on fluorescence emission pattern. We show repre- Figure 7 shows a graphical distribution of the Spectral
sentative spectra in Fig. 6 and characteristics and occurrence Types across different depths. Points of some Spectral
frequencies of these Spectral Types in Table 2 in order of Types have a higher frequency in one depth bin compared
increasing lmax. We used a naming scheme based on position with other bins, which suggests some type of change in
of the lmax to the nearest nm, followed by a descriptor of the deposition history, fractionating mechanism, or in situ
peak shape or pattern position. For example: ‘‘s’’= single peak, modification. For example, Spectral Type L314_s (pink
‘‘d’’= double peak, ‘‘t’’= triple peak, ‘‘m’’=multiple peaks, colored zones in Fig. 7) was mostly found in the upper firn
and ‘‘complex’’ indicating a complex pattern. We designate ice zone, whereas Spectral Type L341_s was mostly found
‘‘sharp’’ to indicate a peak that is narrower in comparison to the in the glacial ice zone (older ice), although both Spectral
other spectral signatures and ‘‘br’’ to indicate lmax peaks that Types were found throughout the borehole column. Of note
appear broader than the other spectral signatures. We also is the Spectral Type L385_d_complex spectral type. This
use ‘‘a’’= asymmetrical for peaks that are not symmetrical Spectral Type was only found in the 80–90 m bin, was
about the lmax. Many of these descriptors are subjective and specifically localized to 83.82–84.94 m, and was found in
based on the current set of spectral signatures. Additional two Point Clouds at that depth: Point Cloud 12 and 14 (these
descriptions and diagnostic spectral features are presented can be seen as the yellow-colored bubbles in Fig. 8).
in Table 2. The spectral response of L385_d_complex appears simi-
We can use the spectral properties of the lmax to infer lar to Spectral Type L385_m; unlike the localized
some properties of the molecular structure and complexity, L385_d_complex, Spectral Type L385_m is found through-
or at least the electronic structure of the fluorescing material out the borehole. The ratios of intensities for peaks in the
(Bhartia et al., 2008). In general, at low excitation wave- spectrum of L385_d_complex are constant, with the signal
GLACIAL ICE EXPLORATION 1195
FIG. 6. Representative examples of normalized spectra of identified Spectral Types grouped from low to high emission
wavelengths. (A) Spectral Types with lmax < 350 nm: L314_d, L314_s, L325_t, L325_s, L341_s, L341_a (Table 3). (B)
Spectral Types with one or two peaks with lmax between 350 and 380 nm: L358_d, L374_sharp. (C) Spectral Types with
multiple peaks with lmax between 340 and 410 nm: L374_m, L385_br_d, L385_m, L385_d_complex. (D) Spectral Types
with lmax > 410 nm: L413_m, L418_d. lmax, lambda max. Color images are available online.
response at 385.3 nm roughly 3· the response at 341.2 nm. In which are found with similar frequencies in both shallow
contrast, peak intensity ratios of corresponding peaks in the firn and deeper glacial ice. Spectral Type L314_s (light
spectrum of Spectral Type L385_m vary. The constant ratio purple zone), Spectral Type L413_m (orange zone), Spectral
between the peaks of L385_d_complex suggests a uniform Type 358_d, Spectral Type 418_d, and Spectral Type 341_s
spectral composition in all the identified points of interest, comprised 15.4%, 8.1%, 7.0%, 6.8%, and 5.5% of all
whereas the varying peak ratios in L385_m spectra could identified points of interest, respectively. From the plot, it
indicate varying compositions among the different points of appears that Spectral Type L418_d is found predominantly
interest of that Spectral Type. From this characterization, we in the lower glacial parts of the borehole; however, below
suggest that although L385_d_complex could be spectrally 90 m, Spectral Type L418_d is only found in two features in
related to L385_m, it is a distinct and localized subpopulation the point cloud dataset, one of which is a large and bright
of features—consistent with (but not diagnostic for) the 43-point feature (entry 77 in Table 2) and the other is a
possibility of this material being an introduced impurity. The relatively dim single-point feature. Spectral Type 358_d is
localization at the depth where previous drilling efforts halted found primarily in the transition zone at 80–90 m, although
is also consistent with an introduced impurity. there are other rare occurrences in firn, but not glacial ice.
During our analysis of Spectral Type occurrence fre- The other Spectral Types listed in Table 3 and shown in
quency, we present frequency numbers in Table 2 that both Fig. 7 (L314_d, L341_a, L374_sharp, L374_m, L385_br_d)
include and exclude the putative impurity Spectral Type are all minor components and combined make up <15% of
L385_d_complex. After excluding Spectral Type L385_d_ all identified points.
complex, most of the selected points from the point cloud
data are Spectral Type L385_m (green zone in Fig. 7),
3.4. Borehole map from point cloud dataset
which makes up almost 25% of all the selected points and is
distributed in both shallow firn and deeper glacial ice. The Figure 8 shows a consolidated view of all the point cloud
second most common is Spectral Type L325_t (blue zone in data to create a partial map of the borehole by point intensity.
Fig. 7), and it makes up 18.8% of all the selected points, This is in a form of cylindrical projection, with rotational
1196
Table 2. List of Spectral Types Identified in the Point Cloud Dataset
Frequency
Percent of Frequency Frequency in firn-glacial Frequency
total features of occurrence in firn zone transition zone in glacial zone
Spectral Total no. (excluding (hit rate (<80 m, hit (80–90 m, hit (>90 m, hit
type name Additional diagnostic spectral features of features L385_d_complex) per 1000)a rate per 1000)a rate per 1000)a rate per 1000)a
L314_d Lambda max 313.7 nm, strong shoulder at 302.6 nm. 19 1.8 (2.4) 0.148 0.182 0.505 0.057
L314_s Lambda max 313.7 nm, little to no shoulder at 302.6 nm. 1270 12.2 (15.9) 0.992 2.707 0.072 0.100
L325_t Lambda max at 324.6 nm, shoulders at 308.1 and 150 14.4 (18.8) 1.172 1.319 0.794 1.154
341.2 nm with diagnostic dips at 319.1 and 335.7 nm.
L325_s Lambda max at 324.6 nm, little to no shoulder at 20 1.9 (2.5) 0.156 0.182 0.289 0.114
313.7 nm.
L341_s Lambda max at 341.2 nm, symmetric peak. 44 4.2 (5.5) 0.344 0.068 0.289 0.527
L341_a Lambda max at 341.2 with sharp rising asymmetric peak; 9 0.9 (1.1) 0.070 0.205 0 0
324.6 nm <20% signal.
L358_d Double peak at 357.7 and 374.3 nm. 56 5.4 (7.0) 0.438 0.091 3.753 0
L374_sharp Single sharp peak at 374.3 or 379.8 nm; usually <10 5 0.5 (0.6) 0.039 0.091 0 0.014
pixels in spatial extent.
L374_m Broad rise to complex pattern with peaks at 341.2, 363.3, 27 2.6 (3.4) 0.211 0.500 0.217 0.028
374.3, and 390.8. Peaks at 374.3 and 390.9 nm nearly
equal.
L385_br_d Broad doublet with lambda maxes at 341.2–385.3 nm; no 28 2.7 (3.5) 0.219 0.091 0 0.342
prominent peaks in between.
L385_m Broad complex with peaks at 341.2, 363.3, 385.3, and 196 18.9 (24.5) 1.531 1.365 2.959 1.354
390.8 nm and a shoulder at 401.9 nm. Peak at 363.3 nm
generally higher than shoulder at 401.9 nm; peak at
374.3 nm absent.
L385_d_complex Broad rise to peaks at 341.2, 363.3, 385.3 nm, and 239 23.0 (0.0) 1.867 0 17.249 0
390.8 nm and shoulder at 401.9 nm. Peaks at 385.3–
401.9 nm are 2 · higher than peaks at 341.2–363.3 nm.
L413_m Broad rise to multiple peak features with lambda max at 65 6.3 (8.1) 0.508 1.092 0.144 0.214
412.9 nm.
L418_d Sharp rise to peaks at 412.9–418.4 and 435–440.5 nm. 54 5.2 (6.8) 0.422 0.205 0.072 0.627
Clear dip from 423.9 to 429.4 nm.
aFrequency refers to the total number of points of interest with that Spectral Type in the entire dataset versus the total number of laser shots.
GLACIAL ICE EXPLORATION 1197
FIG. 7. Frequency of the different spectral types at varying depths. From the surface to the firn-glacial transition zone,
points of Spectral Type L314_s are common, whereas deeper than 90 m, they are rare. In contrast, points of Spectral Type
341_s are more common in glacial ice. Color images are available online.
degrees from magnetic N along the y axis and depth along the of 0.1 mm (100mm) and dimensions of 105 columns · 413
x-axis. The axes in Fig. 8 are not to the same scale; with a rows. This map had a variety of clearly distinguishable
diameter of 11 cm (4.4 inches), the actual borehole circum- bright features. Supplementary Data S4 has bias-corrected
ference is 0.35 m and would not be easily visible at this spectral data for all of Map 1 listed by pixel coordinate.
graphic scale in an undistorted image. Gray crosses show The collected multiband data were visualized and ana-
locations where the laser fired, but no significant return lyzed by ENVI (Harris Geospatial Systems). Figure 9A
fluorescence signal was detected. Colored bubbles show lo- shows an ENVI plot of the data using 412.9, 385.3, and
cations of identified points of interest; the color of the bubble 313.7 nm data, where the stretch of the visualization was set
in the graphic is based on the Spectral Type of the signal, and so that signals 3 · the standard deviation above an average
the size of the bubble is related to the background-subtracted background across the map are rendered visible. In the
signal strength at the lmax. Deviations from straight tracks figure, multiple artifacts appear as ringlets; these are likely
parallel to the y axis show where the instrumented-drill impurities on the window and are observed across multiple
combination rotated during its path down or up the borehole. maps in the same relative position. Figure 9B shows a set of
From initial inspection, the absolute intensity of the peak selected ROIs corresponding to pixels above the limit of
is not affected by the depth. However, as previously noted, detection (LOD = 3 · noise) in at least one of the instrument
there are a few intense points of interest of Spectral Type bands. The ROIs in Fig. 9B are colorized based on their
L418_d found above 80 m, whereas below 90 m they are assigned Spectral Type.
some of the most intense points of interest encountered, Figure 10A–E shows a detailed view of selected spec-
represented by large nested circles in Fig. 8 at *102 m trally diverse ROIs from Map 1, including several of the
depth. If these Spectral Type L418_d points are removed brightest regions. These are shown next to a stacked plot of
(and they may be one single feature since many of the spectra from the indicated region that are above the limit of
signals are consecutive returns from the same point cloud quantification ( = 10· noise) in at least two wavelengths.
transect), then there is a slight trend of a decrease in signal From the plotted spectra, there is minimal spectral variation
intensity for signals in glacial ice. of the individual pixels composing the region—the regions
are spectrally uniform.
3.5. Map analysis
3.6. Analysis of identified ROIs
More than 26 maps were collected spanning depths of 10–
in map from 93.8 m depth
106.7 m. For detailed analysis, we selected Map 1, a
1 · 4 cm map acquired at a lower depth of 93.8 m and a A list of the identified ROIs from Map 1 is presented in
rotation of 53 degrees shown in Fig. 9 (these coordinates are order of increasing depth below the surface in Table 3. The
for the lower right corner of the collected map). The exact spectral characteristics were defined based on similarities
dimensions were 1.05 · 4.13 cm. Map 1 had a shot spacing with the spectral classifications described in Table 2. No
1198 MALASKA ET AL.
FIG. 8. Plot of point cloud data in
borehole coordinates. Cylindrical co-
ordinates with rotation relative to
magnetic north along x-axis, and depth
in borehole along y-axis. Gray crosses
indicate locations where laser data
were acquired but signals above the
noise level were not detected. Colored
bubbles indicate detected features,
with color based on lmax of detected
feature from purple (314 nm) to red
(418 nm). Size of the bubble is based
on background-subtracted signal in-
tensity at the lmax. Color images are
available online.
new Spectral Types were observed during the analysis of the WATSON instrument is up to 2 cm deep in clear
Map 1—all the detected features could be placed in the bubble-free laboratory ice. The exact pathlength inside the
Spectral Type classification scheme presented in Table 2. boreholes is not known, but we assumed roughly <2 cm due
Map 1 ROIs had a large relatively large spectral diversity; of to bubbles; thus, features and connections of extended
the 14 Spectral Types noted in Table 2, 5 of these were features that are deeper than this pathlength would not be
observed in Map 1. Supplementary Data S5 lists all the observed in our map and could create an apparent visual
pixels extracted from all the ROIs in Map 1. disconnection.
Several of the ROIs with similar emission spectra were The areal size of the ROIs ranged from 0.06 to 1.08 mm2,
characterized as separate regions with perceived morpho- with an average area of 0.4 mm2. If ROIs 4–6 were com-
logical breaks with other ROIs. This includes the cluster of bined, they would make an extended region with a spatial
Features 4–6 and Features 14–16 (Fig. 9). Since they are area larger than 2 mm2 and if ROIs 14–16 were combined,
proximal to each other and spectrally related, they could be they would make an extended region with a spatial area
perceived as part of the same feature, if the connections larger than 2.3 mm2 (the extended regions would be larger in
were not visible due to their depth in the ice. In the work of area than the summed individual regions since some ap-
Eshelman et al. (2019), we reported that the pathlength of parently empty pixels lie between the regions).
GLACIAL ICE EXPLORATION 1199
Table 3. List of Features Identified in the Selected Detailed Map
Average Average
Map intensity Maximum Aspect Orientation intensity per
Spectral location Lambda (counts at Area Area dimension ratio clockwise from pixel per
Region ID type (x, y)a max (nm) lambda max) (pixels) (mm2) (mm) (w:h)b horizontalb region area
1 L341_s (4.5, 37.8) 341.2 652 15 0.15 0.81 1.6 -49 43
2 L325_t (3.9, 37.2) 324.6 4693 44 0.44 0.98 1.4 -43 107
3 L385_m (2.9, 34.3) 385.3 1448 24 0.24 0.60 1.4 -90 60
4 L385_m (6.5, 28.7) 385.3 681 30 0.3 1.10 2.0 -47 23
5 L385_m (5.6, 28.1) 385.3 738 90 0.9 1.79 3.1 -58 8
6 L385_m (6.4, 27.9) 385.3 1057 81 0.81 1.46 1.4 -45 13
7 L341_s (10, 26.8) 341.2 2435 20 0.2 0.91 1.8 -42 122
8 L385_m (5.6, 26.3) 385.3 562 11 0.11 0.71 1.6 -45 51
9 L325_s (9, 25.9) 324.6 491 38 0.38 1.31 2.1 -46 13
10 L325_t (8, 26) 324.6 316 36 0.36 1.29 2.9 -85 9
11 L385_m (7.3, 25.6) 385.3 743 39 0.39 1.21 3.4 -78 19
12 L325_t (5.1, 24.9) 324.6 720 6 0.06 0.40 2.0 -90 120
13 L374_sharp (1.4, 10.8) 379.8c 1916 9 0.09 0.40 1.8 -56 213
14 L385_m (5.2, 9.2) 385.3 794 73 0.73 1.53 2.3 -69 11
15 L385_m (4.4, 8) 385.3 588 108 1.08 1.70 1.2 0 5
16 L385_m (4.4, 7.2) 385.3 641 46 0.46 1.28 1.4 -47 14
17 L385_m (5.8, 6.5) 385.3 389 34 0.34 0.84 2.2 70 11
18 L374_m (1.4, 4.1) 374.3 569 16 0.16 0.74 2.5 -77 36
19 L325_t (2.6, 2.6) 324.6 2677 15 0.15 0.50 1.4 -77 178
aMap coordinates are given in mm relative to lower left corner (0, 0) in Fig. 9. Each pixel is 0.1 mm.
bThese values are after spatial adjustment in the z-direction to account for the true orientation and dimensions.
cFor this particular feature, the lambda max was at 379.8 nm, and the intensity at 379.8 nm is recorded here.
The maximum linear dimensions of the ROIs are also The intensities of the signals across different ROI were
indicated; in general, the largest features in Table 3 are less compared by using the background-subtracted signal in-
than 2 mm across. If ROIs 4–6 were combined, they would tensity at relative lmax wavelengths for each region’s as-
make an extended region 2.61 mm across; whereas if 14– signed Spectral Type. The assumption is that regions may
16 were combined, they would make an extended region have similar fluorescence intensities and similar quantum
3.34 mm across. If the extended regions and remaining yield. We determined the average intensity by summing all
isolated ROIs were binned into 1 mm size bins according to the background-subtracted intensities at the lmax, then di-
the maximum dimension, the resulting binning and number viding by the number of pixels for the overall feature. In
of entries would be: 0–1 mm: 10; 1–2 mm: 3, 2–3 mm: 1, Map 1, we found that the pixels with the highest intensity
and 3–4 mm: 1. This is very similar to the distribution were classified as Spectral Type L325_t, specifically,
presented in Fig. 5 for the high-resolution (post-spacing Features 2 and 19.
<0.2 mm) data. Thus, the distribution of the maximum di-
mension of the ROIs in Map 1 appears representative of the
3.7. Estimate of hotspot frequency in Map 1 data
estimated size distribution of the entire point cloud dataset
collected throughout the borehole. The features identified The total number of pixels in view for Map 1 is 43,365
in Map 1 are all at the sub-cm scale. Rare larger features pixels, which allows us to compare the hit rate frequency
may be possible according to the Point Cloud data in Ta- with the point cloud data presented in Table 2.
ble 2 and Fig. 5, however we did not observe these large A summary of data in Map 1 as broken down by Spectral
features in Map 1. Type is in Table 4, with a graphical comparison Map 1
After spatial correction, we measured the maximum Spectral Types to point to cloud Spectral Types presented in
dimensions as well as the maximum dimension orthogonal Fig. 11.
to this direction to determine the aspect ratio of the ROIs; Examination of Table 4 shows that by number of regions,
most of the features had an aspect ratio between 3.4:1 most regions in Map 1 are Spectral Type 385_m. Spectral
(oblong) and 1.2:1 (nearly spherical) with an average as- Type 385_m also has the highest spatial areal coverage, with
pect ratio of 2:1. From the dimension measurements, we more than 72% of the pixels in the ROIs with this spectral
also measured the apparent angle or tilt relative to hori- type, but only slightly more than 50% of the total hotspot
zontal. In general, for most of the ROIs where an angle integrated signals at lmax. Thus, in Map 1, the Spectral Type
could be recorded, regions were tilted roughly 40–70 de- 385_m features are numerous, large, and have low signal
grees counter-clockwise to horizontal, as if they were intensity. The averaged hit rate (pixels of this Spectral type
originally spherical objects that were compressed from the vs. pixels examined) for this spectral type is 12.4 per 1000,
upper left. The sole exception is ROI 17, which has a tilt to which is roughly 10· higher than the overall hit rate deter-
the upper right. mined in Table 2 for this Spectral Type below 90 m depth.
1200 MALASKA ET AL.
In Map 1, only four ROIs of Spectral Type L325_t are
observed. They make up <15% of the total hotspot area yet
have more than 30% of the total signal. In general, these
signals are bright. In Map 1, they have a hit rate frequency
of more than 2 points per 1000 laser shots, but in the point
cloud data in Table 3, they have an average hit rate of 1.2
points per 1000 laser shots for locations below 90 m; thus,
the frequency of these points is only 2 · ‘‘typical’’ expected
values. We should recall that Map 1 was selected for its
richness of ROIs, thus the hit rate is expected to be above
those for bulk point cloud data.
The hit rate for Spectral Type L325_s is enhanced relative
to that observed in the point cloud dataset for depths below
90 m, whereas for L341_s it is roughly similar (Fig. 11). For
L374_sharp and L374_m, the hit rate is higher than the point
cloud data below 90 m; however, these two Spectral Types
are rare and only one ROI for each is observed in the
map. Thus, Map 1 appears to have a slightly enhanced hit
rate for most spectral types, when compared with point
cloud data, with an exceptionally high number (10· ) of
signals of Spectral Type L385_m.
4. Discussion
4.1. Size and distribution
and size of hotspots and features
Numerous (178) features were identified in the point
cloud data. Most of these were single-point features, al-
though the post-spacing in some point clouds was large
enough that a feature as large as 10.6 mm could be detected
as a single point. Some larger features were noted, although
features larger than 15 mm were rare, comprising <4% of
the total number of features. This is in general agreement
with our Map 1 data, where all of the ROIs, even after
combination of neighboring spectrally similar ROIs, would
have dimensions <4 mm. In preliminary analysis of other
map data, we saw no evidence of ROIs larger than 15 mm.
From the map data presented in Fig. 9, the hotspots appear
to have an area of *0.4 mm2 for the brighter hotspots and
they appear as compact <1 mm diameter circular (Features
2, 3) to ring-like (Features 7, 13, 19) features with near-
FIG. 9. Different views of in situ fluorescence data ac- uniform spectral signatures of varying intensity. The aspect
quired from the ice borehole from Map 1 at 93.800 m depth ratio (1.2–3.4:1) and angle of apparent compression in Map
below the surface. (A) Contrast-adjusted stretch of RGB 1 is consistent with a uniform compression amount and di-
(412.9, 385.3, and 313.7 nm). White dotted arrows indicate
rection of compression across the scene that causes any
artifacts, mostly ringlets, likely on the optical window that
appear in multiple maps. (B) Annotated ROIs with colored initially spherical objects to appear slightly out of round, but
squares indicating the 19 brightest features in this scene. not completely flat.
Color key: Blue corresponds to Spectral Type L325_t fea- There was no clear evidence for layers in the Map 1 data.
tures, Cyan is Spectral Type L341_s, Yellow/Orange/Melon The point cloud data only provides one dimension; thus,
is Spectral Type L385_m, Green is Spectral Type series of consecutive points would look similar to a thick
L374_sharp, and Magenta is Spectral L374_m. Feature la- horizontal layer if only the vertical axis were scanned.
bels relate to Table 4 and text. Scale of all scenes as shown Some sections of the borehole were scanned with more
in panel B. The top margin of the map image is toward the than one vertical transect, but at a different angle. In many
top of the borehole. The left and right edges are 41.3 and cases, a track with a hotspot at one rotational angle did not
51.7 mm along the circumference of the borehole from
show a corresponding hotspot at the same depth but a
magnetic N, respectively. The bottom edge of the map is
located at 93.7987 m depth from the surface, whereas the top different rotational angle. This observation is consistent
edge is located at 93.7575 m depth. ROI, region of interest. with discrete spots, rather than layers. As an example,
Color images are available online. Fig. 12 shows a detailed map of in situ WATSON scanning
of a section of the borehole 92.8 to 95.1 m with multiple
point cloud tracks. Detected hotspots in one track are not
repeated on another track. This effect had previously been
noted by Rohde et al. (2008) and Rohde (2010) during
FIG. 10. Detailed view of five se-
lected ROIs from Map 1 taken at
93.8 m depth shown in Fig. 9 and
corresponding extracted spectra in
corrected instrument counts for
region-of-interest pixels with peaks
10 · above interpolated noise. Each
line shows extracted spectra from a
given pixel plotted on an intensity
scale extending to the maximum
fluorescent maxima values for that
ROI. (A) ROI 2, of Spectral Type
L325_t, centered at (3.9, 37.2 mm).
(B) ROI 3, of Spectral Type
L385_m, centered at (2.9, 34.3 mm).
(C) ROI 7, of Spectral Type L341_s,
centered at (10, 26.8 mm). (D) ROI
13, of Spectral Type L374_sharp,
centered at (1.4, 10.8 mm). (E) Fea-
ture 18, of Spectral Type L374_m
centered at (1.4, 4.1 mm). In this
same view, there is a circular region
at lower left that is of Spectral Type
L325_t (spectral data not shown).
Coordinates given in x-offset and
y-offset from origin of Fig. 9 map at
lower left. Scale in all plots as
shown. White arrows indicate se-
lected regions. See Fig. 9 for map
location of these features and Ta-
ble 4 for further details. Color ima-
ges are available online.
1201
1202 MALASKA ET AL.
Table 4. Hit Rate Calculated from Analysis of Map 1
Spectral Number Percent of Areal coverage Areal % of Sum signal at Percent of Hit rate frequency
type regions total hotspots (pixels) total hotspots lambda max (counts) total signal per 1000
L325_t 4 21.1 101 13.7 262,336 34.0 2.329
L325_s 1 5.3 38 5.2 18,668 2.4 0.876
L341_s 2 10.5 35 4.8 58,487 7.6 0.807
L374_sharp 1 5.3 9 1.2 17,247 2.2 0.208
L374_m 1 5.3 16 2.2 9099 1.2 0.369
L385ma 10 52.6 536 72.9 406,504 52.6 12.360
aFeatures 4–6 and 14–16 are not combined in this analysis.
laboratory fluorescence scanning of WAIS Divide and observations, most of these would have to be limited to
GISP2 core sections. It is assumed that the deposition areas that were scanned with only a single track. The data
layers at our borehole would be relatively flat lying, so that presented in Figs. 8, 9, and 12 suggest that fluorescent
any distinct spectral layers should be noted by a series of materials such as microbes and organics are not emplaced in
spectrally similar hotspots at nearly the same depth on evident layers at the fine scales we examined.
different tracks. The image in Fig. 12 has been distorted to With this method and at this fine resolution, the hotspots
better show the rotational dimension (x-axis in figure) of we observed appear to be stochastically distributed. The
the borehole. Given the 11 cm diameter of our borehole, a sizes varied from 74 mm to a single pixel in the point cloud
dip of 45 degrees would generate a 0.1 m deviation from data, and from 0.4 to 3 mm in the map data. In both datasets,
one side of the borehole to the opposite side and should smaller sizes were more prevalent.
still be evident in the graphic. In an analysis of 224 nm excitation fluorescent signals
In general, features do not line up with other features on along a 350mm spaced linear track of an isolated and pre-
parallel tracks, arguing against consistent layers or bands at pared GISP2 ice core (the equivalent of a WATSON point
a given depth. Although we saw no evidence of layers from cloud, but with a different excitation wavelength), Rohde
the Map 1 or point cloud data, it is possible that some of the et al. (2008) showed that both microbes and nonmicrobial
detected broad features in the point cloud data occur in aerosols are deposited onto the ice in bursts that are dis-
layers, zones, or even spotty confined zones, but to fit our continuous on a vertical scale of millimeters to centimeters.
FIG. 11. Comparison of Map 1 Spectral Type frequency with frequencies of point cloud data from varying depths (Fig. 4).
Map 1 has an elevated number of features, although the absolute signal intensities are lower than for the bulk point cloud
data (Supplementary Fig. S2). Color images are available online.
GLACIAL ICE EXPLORATION 1203
FIG. 12. Detailed plot of point
cloud data in borehole coordinates
near 94 m depth. Features do not line
up with other features on parallel
tracks, arguing against a consistent
layer at a given depth. Legend is same
as in Fig. 8. Cylindrical coordinates
with rotation relative to magnetic
north along x-axis, and depth in
borehole along y-axis. Gray lines in-
dicate locations where laser data were
acquired but signals above the noise
level were not detected. Colored bub-
bles indicate detected features, with
color based on lmax of detected fea-
ture from purple (314 nm) to red
(418 nm). Size of the bubble is based
on background-subtracted number of
instrument counts at the lmax. Color
images are available online.
This is consistent with our fine-scale results showing that map data—average hotspot feature size does not appear to
materials are discontinuously distributed both by depth and vary according to depth. These observations suggest that the
in borehole rotation. initial emplacement places a spectrally uniform material at a
specific location and size in the near-surface ice, and that the
transition from firn to glacial ice does not cause significant
4.2. Constraints on distribution
migration or transformation.
and emplacement mechanisms
The observed size of the detected particles constrains the
Our data allow us to evaluate possible emplacement origin and emplacement processes. The maximally estimated
mechanisms for fluorescent materials that explain our ob- size for our largest detected ROI is 74.5 mm (Supplementary
servations. From analysis of our combined point cloud data, Table S2, entry 38, and Supplementary Fig. S1). However, if
the estimated feature size does not appear to vary with depth. the points at the ends of the consecutive series of signals are
Features at the top of the borehole column do not appear estimated to have hit the outer margin of the extended fea-
significantly larger than those at the base, after the firn-glacial ture, then the minimum possible size can also be estimated
transition, suggesting preservation mechanisms in the ice. based on post-spacing; the minimum size of the feature in
This observation from the point cloud dataset is also con- Supplementary Table S2, entry 38 is 63.9 mm. Thus, the
sistent with preliminary examination of our detailed raster estimated size of this feature ranges from 63.9 to 74.5 mm. In
1204 MALASKA ET AL.
general, particulate materials larger than 10 mm (0.01 mm) have fluorescent spectra similar to whole-cell bacterial cul-
will settle in a matter of hours from an air suspension. tures and field isolates as well as similarity to the amino acid,
‘‘Heavy dust’’ is defined as materials <1 mm; some of our tryptophan (Fig. 13B) (Teale and Weber, 1957; Coble, 1996;
detected features are larger than that by nearly two orders of Eshelman et al., 2019). We, thus, take the suite of Spectral
magnitude. Thus, a direct meteoric origin that deposits pure Types with lmax values from 325 to 341 nm as being con-
fluorescent organics directly on the surface at these sizes sistent with microbes or biologically derived materials (in-
seems unlikely. However, an indirect process, such as an cluding biofilms, see Smith et al., 2016), lignin-breakdown
enriched single snowflake with fluorescent material of this products, and bicyclic aromatic species. This suite of Spectral
size, is possible. Types is not consistent with simple benzenes that have
A more likely scenario is that organic material is origi- shorter fluorescence lmax values <290 nm or higher-order
nally deposited, perhaps as diffuse regions, but is then multi-ring aromatic systems that have longer lmax values
concentrated into discrete and stochastically located hot- (e.g., phenanthrene) (Fig. 13C). Shorter-wavelength fluo-
spots in the ice at a fine scale. This would happen at the rescing Spectral Types, such as L314_d and L314_s (data
upper surfaces of the ice, as the size of the regions and shown in Fig. 6A), are similar to fluorescence features seen in
spectral nature is preserved intact in the ice column. The EEMs data from WAIS Divide ice cores that were ascribed to
uniform size may be related to diffusion through the ice, simple lignin phenols (D’Andrilli et al., 2017; see low ex-
with diffusion occurring quickly at the surface to form an citation and emission wavelength fluorescence example C1
initial ca. 1 mm spot, which is then ‘‘locked in’’ and carried presented in Fig. 2A in that text.).
through the firn to glacial ice transition and densification.
This may occur at a fine scale that has not been previously 4.3.2. Longer wavelength signals. The Spectral Types
observed by bulk analysis. Exact mechanisms are con- at longer wavelengths >>341 nm are consistent with PAHs
strained by spectral uniformity, scale that favors smaller or larger aromatic species (see Fig. 14 for molecular
hotspots (1 mm being common), favorable energetic re- structures) and are consistent with spectral signatures that
quirements, and stochastic emplacement. were observed by Rohde et al. (2008). These chemical
species have been categorized as more complex compared
4.3. Compositional constraints: with chemical species that fluoresce at shorter wavelengths
comparison with laboratory spectra from ancient and modern ice cores (D’Andrilli et al., 2017)
and can originate from microbial and continental nonmi-
Spectral Types detected in the Summit borehole are
crobial sources (e.g., Pony Lake and Suwannee River ful-
compared with selected laboratory standards of microbial,
vic acids).
chemical, and field sample materials in Fig. 13. This figure
In general, the emission maxima of Spectral Type L358_d
also compares our data with excitation–emission data in the
and L374_sharp are above expected values from microbial
literature when a 250 nm excitation wavelength data could
cells where the spectra may be driven by fluorescence from
be isolated (Coble et al., 1990, 2014; D’Andrilli et al., 2013,
the amino acid tryptophan and other aromatic amino acids
2017, 2020) and includes comparisons with natural organic
(Coble et al., 2014). Instead, the spectral responses of these
matter mixtures that have a more characteristically complex
longer-wavelength Spectral Types may be derived from
fluorescent nature. It is important to note that the standard
more complex materials that may or may not be directly
spectra were acquired at room temperature in aqueous or
related to microbial processes. The spectral features of
methanolic solution in the laboratory, whereas WATSON
-  Spectral Type L358_d (Fig. 13C) and the sharp Spectralspectra were acquired in an ice matrix at roughly 20 C in
Type L374_sharp (Fig. 13D) are both similar to spectra from
field conditions; consequently, our measured signals may
PAHs such as phenanthrene, pyrene, and 1-methylpyrene.
not overlap exact regions of laboratory standards. Although
Figure 13E shows the complex spectrum of Spectral Type
a rigorous identification is beyond the scope of this article,
L385_m and L374_m as solid lines, and comparison spectra
our identified spectra compare favorably with likely material
from PAHs already listed including three-ring linear-fused
targets in the polar environment and with previous studies of
anthracene.
fluorescent organic matter of various sources (Coble et al.,
Although none of the spectra are exact matches, alkyl or
2014; D’Andrilli et al., 2017).
heteroatom-substituent substitution as well as changes of
matrix conditions (water vs. ice, temperature effects) could
4.3.1. Shorter wavelength signals. Spectral Types cause changes to the reference spectra ( Johnson et al.,
L325_t and L325_s have lmax values around 324.6 nm and 2011). Thus, we can state that the spectral signals are
are similar to whole-cell spectra of bacterial psychrotolerant broadly consistent with fluorescence spectra from PAHs,
strains such as Shewanella oneidensis MR1 and Bacillus and that they are not similar to the fluorescent signals of
pumilis (shown in Fig. 13B). Materials with peaks in this microbial cells.
range also include broken-down products of lignins such as Sources of PAHs for the Greenland ice sheet include the
lignin phenols (Hernes et al., 2009; Coble et al., 2014); these burning and emissions coming from Northern Hemisphere
signatures are similar to those found in glacial Antarctic ice sources ( Jaffrezo et al., 1994; Slater et al., 2002; Von
WAIS Divide ice cores (D’Andrilli et al., 2017). The mo- Schneidemesser et al., 2008), including boreal forest fires
lecular structures of vanillic acid (an example of a lignin and fossil fuel combustion. Figure 13F shows spectra from
phenol) as well as other organic molecules used for spectral Spectral Types L413_m and L418_d. These are compared
comparison are shown in Fig. 14. Naphthalene, a two-ring with Pony Lake fulvic acid, Suwannee River fulvic acid, as
PAH, also has a similar spectrum with an emission lmax in the well as fluoranthene, a PAH with a chemical structure
same region. The features labeled L341_s and L341_a both similar to naphthalene fused to a benzene ring via a central
GLACIAL ICE EXPLORATION 1205
FIG. 13. Comparison of identified features with known biological, chemical, and environmental materials. In the
figures, spectra isolated from WATSON borehole scanning are in bold lines, whereas laboratory standards are in dotted or
dashed lines. (A) Spectral Type L325_t (bold black line) and L325_s (bold blue line) with Shewanella oneidensis MRI
whole cells, naphthalene, vanillic acid, and p-benzoquinone. (B) Spectral Type L341_s (bold black line) and L341_a
(bold light blue line) with Bacillus pumillus, Antarctic bacterium isolate (Dieser et al., 2019), and tryptophan. (C)
Spectral Type L358_d (bold black line) with phenanthrene. (D) Spectral Type L374_sharp (bold black line) with methyl-
1-pyrene and pyrene. (E) Spectral Type L385_m (bold black line) and Spectral Type L374_m (bold green line) with
naphthalene, phenanthrene, pyrene, and anthracene. (F) Spectral Type L413_m (bold black line) and Spectral Type
L418_d (bold red line) with Pony Lake and Suwannee River fulvic acid samples (D’Andrilli et al., 2013), fluoranthene,
and perylene standards. Color images are available online.
five-membered-ring. The PAHs have been detected in sample is derived from polar microbial sources devoid of
shallow snow pits at Summit Station by multiple researchers higher-order plants, whereas the Suwannee River fulvic
( Jaffrezo et al., 1994; Slater et al., 2002; Von Schneide- acids are largely influenced by continental higher-order
messer et al., 2008), so the presence of potential PAH plants. Since Greenland is surrounded by continents with
chemical signals is consistent with known literature. There extensive boreal forests, it is likely that the Greenland ice
is similar overlap of Spectral Type L413_m and the Pony sheet receives continental organic materials that may be
Lake fulvic acid sample, suggesting that materials deposited preserved in firn and ice layers. Therefore, a continental
in the ice may be similarly complex. The Pony Lake fulvic higher-order organic end member sample, such as the
1206 MALASKA ET AL.
FIG. 14. Molecular structures of various laboratory chemicals used as comparison spectra organized in order (left to right,
top to bottom) of increasing electronic delocalization and longer lmax wavelength.
Suwannee River fulvic acid, is a reasonable environmental surface, and finally in the silt-infused bottom ice near the 3
comparative sample for this work. km-deep base of the ice sheet, the cell counts went from 106
Examples in Fig. 13 show that the Greenland ice sheet to 109 cells/cm3 (Miteva et al., 2009). In the near-surface
contains fluorescent signals over short and long wave- ice, previous work by Yung et al. (2007) with the GISP2 ice
lengths, which suggests that the materials deposited and core found 7 · 103 microbial cells/cm3 with nearly 400 en-
preserved in ice vary in complex nature and may indicate dospores/cm3 at a depth of 93.96–94.21 m, which corre-
diverse sources. Although similar spectral signals to labo- sponded to an age of 295 years. Our field site was located
ratory standards or environmental mixtures do not confirm only 7 km distant from the GISP2 location on the flat surface
or quantify its presence, it suggests a similar chemical na- of the top of the ice sheet but several years later, these
ture. In summary, all of our detected spectral features from depths roughly correspond within 10 meters to our targeted
the Summit borehole acquired by WATSON are similar to borehole column.
our measured laboratory microbial, chemical, and environ- We can use the microbial concentrations from the lit-
mental material comparison samples that would be expected erature to speculatively estimate our instrument response
to be derived from microbes, pollution, aerosols, and eolian based on our measured signal (Bhartia et al., 2010). Our
dusts delivered to the ice cap surface. Map 1 was taken at a depth of 93.8 m: thus, from previous
analyses (Yung et al., 2007; Miteva et al., 2009) we should
expect roughly 104 cells cm3 to be present. If we take the
4.4. Estimates of instrument sensitivity
Map 1 spatial dimensions to be roughly 1 · 4 cm and as-
Using some speculative assumptions, we will try to esti- sume an optical penetration depth of 1 cm (this value is
mate our instrument sensitivity based on literature concen- conservatively decreased from the Eshelman et al., 2019
trations of materials from Summit Station ice samples. Ice estimate of 2 cm), then the entire ice volume scanned in
cores from the Greenland Ice Sheet coring project (GISP2) Map 1 is 1 · 4 · 1 cm = 4 cm3. If Map 1 is representative of
borehole at Summit Station provided a depth transect from literature cell concentrations (Yung et al., 2007; Miteva
the surface down to the base of the Greenland Ice Sheet ice et al., 2009), we would thus expect 4 · 104 cells to be
sheet a little more than 3 km below the surface. For mi- present in the Map 1 scanned volume. For the purposes of
crobial concentrations, analysis of the GISP2 ice cores our estimate, we postulate that Spectral Types L325_t,
demonstrated that the cell counts varied from 104 to 105 L325_s, and L341_s are all microbial in nature, and that
cells/cm3 from the surface to a depth of 1.5 km below the quantum yields are roughly similar. We, thus, estimate that
surface, to 104–106 cells/cm3 from 1.5 to 2.5 km below the the integrated lmax signal intensity in Map 1 for these
GLACIAL ICE EXPLORATION 1207
combined Spectral Types of 341,085 instrument counts of PAH in one pixel of the entire map scanned area. This is
corresponds to 4 · 104 cells. This provides an estimated approximately equivalent to the highest reported sensitiv-
correspondence of *8.5 counts per cell. Assuming a LOD ity for a laboratory method (ca. 0.1 ng/kg1 snow) (Von
for these wavelengths of 300 instrument counts, this means Schneidemesser et al., 2008).
it might be possible to detect a minimum of 35 cells in our These speculative estimates suggest that our in situ instru-
sample if they are all in the same pixel of microbial con- ment has the potential to be highly sensitive and is able to detect
firmed signals. This estimated value is within the same both microbial and chemical signals at low native concentra-
order of magnitude to the value presented in the work of tions in a field setting at finer scales than can be measured by
Eshelman et al. (2019) based on laboratory measurements conventional ice core extraction and laboratory melting.
of Escherichia coli in ice. At an estimated 8.5 counts per cell,
this implies that Feature 2 in Map 1, with 206,492 integrated 4.5. Prediction of hotspot analysis
counts, would correspond to roughly 24,000 cells spread over versus bulk dilution analysis
an interrogation volume of 0.004 cm3 (0.004 cm2 spatial ar-
ea· 1 cm estimated optical depth), which is a local concen- A future deep drilling mission through the icy crust of an
tration of 5.5· 106 cells cm3—more than a 500· increase Ocean World may sample melted ice during its journey on
compared with the estimated bulk concentration. the way to the ocean, or it could examine hotspots embed-
We can use the same logic to speculate on the instrument ded in he borehole walls: Which is better? This is the same
response to longer wavelength signals. We speculate that the question as comparing traditional bulk melt analysis with
longer wavelength signals are due to PAHs. The PAHs have been our fine resolution scanning. During traditional melt analy-
detected in shallow trenches at Summit Station by multiple re- sis, any contained fluorescent microbial or chemical hot-
search groups (Jaffrezo et al., 1994; Slater et al., 2002; Von spots are diluted into the volume melted and subsequently
Schneidemesser et al., 2008). The detected concentrations of analyzed. We can examine the effect of dilution by using
PAHs in snow are low—on the order of a few nanograms per our collected data from the point clouds and Map 1 to
kilogram of bulk snow; past measurements required processing compare detection levels between typical hotspots in the
of large amounts of snow and highly sensitive detection tech- point cloud and map and a scenario if the entire datasets
niques in the laboratory (Von Schneidemesser et al., 2008). Of (point clouds and Map 1, separately) had been subjected to
the PAHs detected, roughly 30% was fluoranthene, 20–30% was bulk melt analysis. To do this, we integrated the signals for
phenanthrene, 20–30% was pyrene, and 15–20% was naphtha- all the point cloud and map points of interest and ROIs,
lene, with benzo[a]pyrene, benzo[e]pyrene, and benzo[ghi]per- respectively, and averaged them across all the hotspot pixels
ylene making up about 5–10% each (Jaffrezo et al., 1994; Slater in each dataset to create an ‘‘average’’ hotspot spectrum.
et al., 2002) (for molecular structures of the base ring systems, The averaged spectra per pixel for all the points of interest
see Fig. 14). It was noted that benzo[a]pyrene decomposes from the point cloud dataset are shown as a dot-dash brown
quickly in ice as evidenced by lower amounts detected from line in Fig. 15, whereas the averaged ROIs spectrum per
lower depths of a shallow snow trench, whereas fluoranthene and pixel for Map 1 data is the green line. For the averaged point
pyrene undergo moderate rates of decomposition in ice (Jaffrezo cloud data, the spectrum has more of a response compared
et al., 1994). Phenanthrene is stable and would be expected to with Map 1 averaged spectrum at wavelengths above
dominate after degradation of the other products. (The degra- 400 nm; this is due to points of interest from Spectral Type
dation pathways and fluorescence characteristics of the degra- 413_m and Spectral Type L418_d. These two Spectral
dation products were not determined.). Types are not observed in Map 1 data. Next, we took the
Even though it is known to degrade moderately, we will summed signal intensity for the points of interest and ROIs
speculate that the L374_sharp signal is from pyrene due to and diluted them across all the point cloud points (127,799
its spectral similarity (although there are differences be- pixels) and the Map 1 points (43,365 pixels), respectively.
tween the method of collecting both spectra that preclude a This represents the scenario where the integrated hotspots
positive identification). We will assume that pyrene is have been diluted by traditional bulk analysis. This exercise
present in the Map 1 data at a concentration of 4 ng/kg resulted in a signal dilution or decreased signal of 123· less
snow = ca. 8 ng/kg ice [this is the highest concentration and 59 · less, for the point cloud and Map 1 data, respec-
reported in the work of Slater et al. (2002), as a conser- tively. The diluted spectra are shown as the double dot-
vative figure we are assuming a 2 · compaction to glacial dashed thin green line and the thin red line in Fig. 15 for the
ice]. As stated earlier, assuming a conservative optical diluted point cloud data and the diluted Map 1 data, re-
penetration depth of 1 cm, the entire ice volume scanned in spectively. At this scale, the two lines are close to baseline
Map1 is 1 · 4 · 1 cm = 4 cm3. Since our measured density and lie nearly on top of each other. A slight increase in
was 0.84 g/cm3, we would expect 7 pg/cm3 of pyrene in our signal intensity at wavelengths above 400 nm for the diluted
ice sample. For the volume in Map 1, this would be a total point cloud data can barely be discerned.
of roughly 28 pg present in Map 1’s interrogated volume. In the same figure, we also plot the average LOD across the
We will use Spectral Type L374_sharp feature (ROI 13 in map area based on the average extracted backgrounds. We took
Fig. 8 and Table 4) as the sole indicator of pyrene in that all the local background spectra used for feature background
volume to estimate the levels of detection. In Map 1, ROI subtraction and used the average standard deviation of the
13 was the sole representative of Spectral Type L374_sharp background signals. These were multiplied by a factor of 3 to
and had a combined signal of 17,244 instrument counts at generate the LOD (shown as a hashed black line in the figure).
the lmax. This means that 616 counts of Spectral Type Any spectra falling below this limit would be not considered
L374_sharp are the equivalent of 1 pg of pyrene. Assuming significant.Since all the dilutedsignals fromeither the point cloud
an LOD of 200 counts, our equivalent LOD is thus 0.33 pg or map pixels fall under this limit, they would not be detected.
1208 MALASKA ET AL.
FIG. 15. Plots showing average spectra. Dash-dotted red line shows the averaged spectra by combining all the point cloud
points of interest and dividing by the number of point cloud hot spot pixels. Green line shows average spectra by combining
all the map feature pixels together and dividing by the number of map feature pixels. The lower thin red line (just above
zero) shows the integrated point cloud feature pixels divided by all the interrogated points in the point cloud. The lower
green dot-dash line shows the average spectra across all the pixels in the map, including those without features. The shaded
dash black line shows the average LOD (3 · noise) of the entire map area—it is roughly equivalent to a typical point cloud
LOD. After full dilution, the diluted spectral signal is well below the LOD level and would not be detected. LOD, limit of
detection. Color images are available online.
This has important implications for future instrumentation few larger features measuring up to 60 mm in dimension. We
and demonstrates the utility of detecting hotspots at fine did not see any of these larger features in our raster maps,
scales. By examining hotspots concentrated in borehole ice instead noting that most features were <2 mm in size. We
rather than bulk melt dilution, we can gain 50–100· in signal found Spectral Type-dependent differences in hit rate fre-
intensity. Depending on the instrument threshold, this could be quencies as well as differences in frequencies above and below
the difference between detecting and not detecting a signal. the firn-glacial ice transition. The Spectral Types we identified
In situ ice scanning also maintains spatial integrity and allows had spectra that are similar to microbes, lignin-phenols, fused-
more detailed analysis on associations, layering (or lack ring aromatic molecules, including PAHs, and environmentally
thereof), and spot distribution. This analysis, thus, provides complex fulvic acids of microbial and multicellular origin.
impetus for future in situ scanning and spatial mapping of ice We found no evidence of fluorescent material emplaced
boreholes compared with traditional bulk analysis. in layers, at least at the fine scales examined. There was no
apparent correlation between the size of the feature and its
depth. Both map and point cloud data had features of
5. Conclusions
roughly the same distribution in shallow firn and in deeper
Our DUV fluorescence mapping spectrometer revealed di- glacial ice. This suggests that the mechanisms of emplace-
verse spectral features from 0 to 107 m in both firn and glacial ment favor smaller features and may occur at shallow depth,
ice. We used both a linear point cloud scanning mode and a and then ‘‘lock in’’ early in the process. The distribution of
raster mapping mode with a coupled drill-instrument combi- most spectral types appears the same in glacial as for firn
nation. Both instrument modes detected spectrally diverse ice, although some exceptions were noted.
features consistent with individual, stochastically distributed We successfully demonstrated in situ detection and spectral
<20 mm globules of material. Overall, the frequency of the analysis of diverse punctate chemical signals similar to microbial
hotspot intensities followed a power law, with intense signals and other organic molecules in a drilled ice borehole at Summit
only a few percent of the detected signals. From point cloud Station, Greenland. Data obtained from Summit Station show
analysis, we found that most of the features were small with a that our DUV fluorescence mapping spectrometer with 100mm
GLACIAL ICE EXPLORATION 1209
spacing can spatially resolve spectrally unique features, whereas Bhartia R, Salas E, Hug W, et al. (2010) Label-free bacterial
traditional bulk melt analysis removes spatial context and po- imaging with deep-UV laser induced native fluorescence.
tentially dilutes the signal below instrumentation limits of de- Appl Environ Microbiol 76:7321–7237.
tection. Our DUV-based technique can be useful for localizing Boetius A, Anesio AM, Deming JW, et al. (2015) Microbial
the emplacement of organic material in icy environments on ecology of the cryosphere: sea ice and glacial habitats. Nat
Earth. This technique could also be key to the detection of as- Rev Microbiol 13:677–690.
trobiologically relevant organic molecules for a future mission to Campen KR, Sowers TA, and Alley RB (2003) Evidence of
the icy crusts of the Ocean Worlds such as Europa, Enceladus, microbial consortia metabolizing with a low-latitude moun-
and Titan, or perhaps the polar caps of Mars. tain glacier. Geology 31:231–234.
Christner BC, Mikucki JA, Foreman CM, et al. (2005) Glacial
Acknowledgments ice cores: a model system for developing extraterrestrial de-
contamination protocols. Icarus 174:572–584.
The authors would like to thank the entire staff of Summit Coble PG (1996) Characterization of marine and terrestrial
Station for all their support. They would also like to thank CH2M DOM in seawater using excitation emission matrix spec-
HILL Polar Services and the National Science Foundation for troscopy. Mar Chem 51:325–346.
their support during the field season. They are grateful to C.M. Coble PG, Green SA, Blough NV, et al. (1990) Characterization
Foreman for the use of laboratory instrumentation at Montana of dissolved organic matter in the Black Sea by fluorescence
State University and isolated Antarctic bacterium sample from spectroscopy. Nature 348:432–435.
the Cotton Glacier Stream in West Antarctica (Dieser et al., 2019) Coble PG, Lead J, Baker A, et al. (2014) Aquatic Organic Matter
and to D. Austin Nordman at the Jet Propulsion Laboratory/ Fluorescence. Cambridge University Press, Cambridge, pp 35–74.
California Institute of Technology for assistance in instrument D’Andrilli J, Foreman CM, Marshall AG, and McKnight DM
output conversion. They would also like to thank two anonymous (2013) Characterization of IHSS Pony Lake fulvic acid dis-
reviewers for their helpful comments. Copyright 2020. All rights solved organic matter by electrospray ionization Fourier
reserved. transform ion cyclotron resonance mass spectrometry and
fluorescence spectroscopy. Organic Geochemistry 65:19–28.
Author Disclosure Statement https://doi.org/10.1016/j.orggeochem.2013.09.013
D’Andrilli J, Foreman CM, Sigl M, et al. (2017) A 21,000-year
No competing financial interests exist.
record of fluorescent organic matter markers in the WAIS
Funding Information Divide ice core. Clim Past 13:533–544.
D’Andrilli J and McConnell JR (2020) Polar ice core organic
This work was carried out in part at the Jet Propulsion matter signatures reveal past atmospheric carbon composition
Laboratory, California Institute of Technology under the and spatial trends across ancient and modern timescales.
NASA Planetary Science and Technology Through Analog Journal of Glaciology: In Review.
Research (PSTAR) program (NNH14ZDA001NPSTAR). Dieser M, Smith HJ, Ramaraj T, et al. (2019) Janthinobacterium
Government sponsorship is acknowledged. CG23_2: comparative genome analysis reveals enhanced
environmental sensing and transcriptional regulation for ad-
Supplementary Data aptation to life in an Antarctic supraglacial stream. Micro-
Supplementary Data S1 organisms 454:1–18.
Eshelman E, Malaska MJ, Manatt KS, et al. (2019) WATSON:
Supplementary Data S2
in situ detection in subsurface ice using deep-UV fluores-
Supplementary Data S3
cence spectroscopy. Astrobiology 19:771–784.
Supplementary Data S4
Gabrielli P and Vallelonga P (2015) Contaminant records in ice
Supplementary Data S5 cores. In Environmental Contaminants, Developments in
Supplementary Figure S1 Paleoenvironmental Research, Vol. 18, edited by JM Blais,
Supplementary Figure S2 M Rosen, and JP Smol, Springer, Dordrecht. pp 393–430.
Supplementary Table S1 DOI: 10.1007/978-94-017-9541-8_14.
Supplementary Table S2 Garcia-Lopez E and Cid C (2017) Glaciers and ice sheets as
analog environments of potentially habitable icy worlds.
References
Front Microbiol 8:1407.
Baker I (2019) Microstructural characterization of snow, firn Geyer R, Jambeck JR, and Law KL (2017) Production, use, and
and ice. Philos Trans R Soc A 377:20180162. fate of all plastics ever made. Sci Adv 3:e1700782.
Barletta RE and Roe CH (2012) Chemical analysis of ice vein Hernes PJ, Bergamaschi BA, Eckard RS, et al. (2009)
m-environments. Polar Record 48:334–341. Fluorescence-based proxies for lignin in freshwater dissolved
Barletta RW, Pricsu JC, Mader HM, et al. (2012) Chemical analysis organic matter. J Geophys Res 114:G00F03.
of ice vein m-environments II—analysis of glacial samples from Jaffrezo JL, Clain MP, and Masclet P (1994) Polycyclic aro-
Greenland and the Antarctic. J Glaciol 58:1109–1118. matic hydrocarbons in the polar ice of Greenland. Geo-
Bender M, Sowers T, and Brook E (1997) Gases in ice cores. chemical use of these atmospheric tracers. Atmos Environ 28:
Proc Natl Acad Sci U S A 94:8343–8349. 1139–1145.
Bergmann M, Mützel S, Primke S, et al. (2019) White and Johnson PV, Hodyss R, Boler DK, et al. (2011) Ultraviolet-
wonderful? Microplastics prevail in snow from the Alps to stimulated fluorescence and phosphorescence of aromatic
the Arctic. Sci Adv 5:eaax1157. hydrocarbons in water ice. Astrobiology 11:151–156.
Bhartia R, Hug WF, Salas EC, et al. (2008) Classification of Junge K, Krembs C, Deming J, et al. (2001) A microscopic
organic and biological materials with deep ultraviolet exci- approach to investigate bacteria under in situ conditions in
tation. Appl Spectrosc 62:1070–1077. sea-ice samples. Ann Glaciol 33:304–310.
1210 MALASKA ET AL.
Kawamura K, Sukuki I, Fujii Y, et al. (1994) Ice core record of Priscu JC, Fritsen CH, Adams EE, et al. (1998) Perennial
polycyclic aromatic hydrocarbons over the past 400 years. Antarctic lake ice: an oasis for life in a polar desert. Science
Naturwissenschaften 81:402–505. 280:2095–2098.
Liu, Y, Priscu JC, Yao T, et al. (2016) Bacterial responses to Priscu JC, Adams EE, Lyons MA, et al. (1999) Geomicro-
environmental change on the Tibetan Plateau over the past biology of sub-glacial ice above Lake Vostok, Antarctica.
half century. Environmental Microbiology, 18(6):1930–1941. Science 286:2141–2144.
doi: 10.1111/1462-2920.13115. Priscu JC, Christner BC, Foreman CM, et al. (2007) Biological
Liu Y, Priscu JC, Yao T, et al. (2018) Culturable bacteria iso- material in ice cores. In Encylopedia of Quaternary Sciences,
lated from seven high-altitude ice cores on the Tibetan Pla- Volume 2, edited by SA Elias, Elsevier B.V., Cambridge,
teau. J Glaciol 65:29–28. UK, pp 1156–1166.
Lomonaco R, Albert M, and Baker I (2011) Microstructural Rohde RA (2010) The development and use of the Berkeley
evolution of fine-grained layers through the firn column at fluorescence spectrometer to characterize microbial content
Summit, Greenland. J Glaciol 57:755–762. and detect volcanic ash in glacial ice. PhD thesis, University
Mader HM, Pettitt ME, Wadham J, et al. (2006) Subsurface ice of California, Berkeley, Berkeley, CA.
as a microbial habitat. Geology 34:169–172. Rohde RA and Price BP (2007) Diffusion-controlled metabolism
Margesin R and Collins T (2019) Microbial ecology of the for long-term survival of single isolated microorganisms trapped
cryosphere (glacial and permafrost habitats): current knowl- within ice crystals. Proc Natl Acad Sci U S A 104:16592–16597.
edge. Appl Microbiol Biotechnol 103:2537–2549. Rohde RA, Price PB, Bay RC, et al. (2008) In situ microbial
Margesin R and Miteva V (2011) Diversity and ecology of metabolism as a cause of gas anomalies in ice. Proc Natl
psychrophilic microorganisms. Res Microbiol 162:346– Acad Sci U S A 105:8667–8672.
361. Santibáñez PA, Maselli OJ, Greenwood MC, et al. (2018)
Marion GE, Fritsen CH, Eicken H, et al. (2003) The search for Prokaryotes in the WAIS Divide ice core reflect source and
life on Europa: limiting environmental factors, potential transport changes between Last Glacial Maximum and the
habitats, and Earth analogues. Astrobiology 3:785–811. early Holocene. Glob Change Biol 24:2182–2197.
Mayewski PA (1999) GISP2 Ions: B Core (0-200 Meters). Santibáñez PA, Michaud AB, Vick Majors TJ, et al. (2019)
PANGAEA. Available online at https://doi.org/10.1594/ Differential incorporation of bacteria, organic matter, and
PANGAEA.55536 (accessed November 9, 2019). inorganic ions into lake ice during ice formation. J Geophys
McGrath D, Colgan W, Bayou N, et al. (2013) Recent warming Res Biogeosci 124:585–600.
at Summit, Greenland: global context and implications. Skidmore ML, Foght JM, and Sharp MJ (2000) Microbial life
Geophys Res Lett 40:2091–2096. beneath a high arctic glacier. Appl Environ Microbiol 66:
Meese DA, Gow AJ, Grootes P, et al. (1994) The accumulation 3214–3220.
record from the GISP2 core as an indicator of climate change Slater JF, Currie LA, Dibb JE, et al. (2002) Distinguishing the
throughout the Holocene. Science 266:1680–1682. relative contribution of fossil fuel and biomass combustion
Mellerowicz B, Zacny K, Paulsen G, et al. (2018) Development aerosols deposited at Summit, Greenland through isotopic
of a deep drill system with integrated UV/Raman spectrom- and molecular characterization of insoluble carbon. Atmos
eter for Mars and Europa. In AIAA SPACE and Astronautics Environ 36:4463–4477.
Forum and Exposition, Orlando, FL. Smith HJ, Schmit A, Foster R, et al. (2016) Biofilms on glacial
Miteva V (2008) Bacteria in snow and glacier ice. In Psy- surfaces: hotspots for biological activity. NPJ Biofilms Mi-
chrophiles: From Biodiversity to Biotechnology, edited by crobiomes 2:16008.
R Margesin, F Schinner, JC Marx, et al., Springer, Berlin, Summit Guide (2019) Prepared by Polar Field Services for
Heidelberg. pp 31–50. CH2M HILL Polar Services. Available online at http://
Miteva VI and Brenchley JE (2005) Detection and isolation cpspolar.com Accessed April 23, 2019.
of ultrasmall microorganisms from a 120,000-year-old Teale FW and Weber G (1957) Ultraviolet fluorescence of the
Greenland glacier ice core. Appl Environ Microbiol 71: aromatic amino acids. Biochem J 65:476–482.
7806–7818. Tung HC, Bramall NE, and Price PB (2005) Microbial origin of
Miteva V, Teacher C, Sowers T, et al. (2009) Comparison of the excess methane in glacial ice and implications for life on
microbial diversity at different depths of the GISP2 Green- Mars. Proc Natl Acad Sci U S A 102:18292–18926.
land ice core in relationship to deposition climates. Environ Vance SD, Panning MP, Stähler S, et al. (2018) Geophysical
Microbiol 11:640–656. investigations of habitability in ice-covered ocean worlds. J
Pinto B, Maccario L, Dommergue A, et al. (2019) Do organic Geophys Res Planets 123:180–205.
substrates drive microbial community interactions in arctic Von Schneidemesser E, Schauer JJ, Shafer MM, et al. (2008) A
snow? Front Microbiol 10:2492. method for the analysis of ultra-trace levels of semi-volatile
Price PB (2000) A habitat for psychrophiles in deep Antarctic and non-volatile organic compounds in snow and application
ice. Proc Natl Acad Sci U S A 97:1247–1251. to a Greenland snow pit. Polar Sci 2:251–266.
Price PB (2006) Microbial life in glacial ice and implications Whitman WB, Coleman DD, and Wiebe WJ (1998) Prokaryotes:
for a cold origin of life. FEMS Microbiol Ecol 59:217–231. the unseen majority. Proc Natl Acad Sci U S A 95:6578–6583.
Price PB (2009) Microbial genesis, life and death in glacial ice. Yung PT, Shafaat HS, Connon SA, et al. (2007) Quantification
Can J Microbiol 55:1–11. of viable endospores from a Greenland ice core. FEMS Mi-
Priscu J and Christner B (2004) Earth’s icy biosphere. In Mi- crobiol Ecol 59:300–306.
crobial Diversity and Bioprospecting, Chapter 13, edited by Zacny K, Shara M, Paulsen G, et al. (2016) Development of a
AT Bull, Washington, DC: ASM Press. pp 130–145. DOI: planetary deep drill. In Earth and Space 2016, 15th Biennial
10.1128/9781555817770. ASCE Conference on Engineering, Science, Construction, and
Priscu J and Hand KO (2012) The microbial habitability of ex- Operations in Challenging Environments pp 256–266. doi:
traterrestrial icy worlds: a view from Earth. Microbe 7:167–172. 10.1061/9780784479971.027.
GLACIAL ICE EXPLORATION 1211
Address correspondence to:
Abbreviations Used
Michael J. Malaska
Jet Propulsion Laboratory/ lmax ¼ lambda max
California Institute of Technology DUV¼ deep-ultraviolet
4800 Oak Grove Drive EEMs¼ excitation–emission matrices
Mail Stop 183-301 GISP2¼Greenland Ice Sheet Project 2
Pasadena, CA 91109–8099 LOD¼ limit of detection
PAH¼ polycyclic aromatic hydrocarbon
USA
PMT¼ photomultiplier tube
E-mail: michael.j.malaska@jpl.nasa.gov ROIs¼ regions of interest
WAIS¼West Antarctic Ice Sheet
Submitted 29 January 2020 WATSON¼Wireline Analysis Tool for the Subsurface
Accepted 8 May 2020 Observation of Northern ice sheets
Associate Editor: Victor Parro
This article has been cited by:
1. Juliana D'Andrilli, Joseph R. McConnell. 2021. Polar ice core organic matter signatures reveal past atmospheric carbon
composition and spatial trends across ancient and modern timescales. Journal of Glaciology 67:266, 1028-1042. [Crossref]
2. Rohit Bhartia, Luther W. Beegle, Lauren DeFlores, William Abbey, Joseph Razzell Hollis, Kyle Uckert, Brian Monacelli, Kenneth
S. Edgett, Megan R. Kennedy, Margarite Sylvia, David Aldrich, Mark Anderson, Sanford A. Asher, Zachary Bailey, Kerry Boyd,
Aaron S. Burton, Michael Caffrey, Michael J. Calaway, Robert Calvet, Bruce Cameron, Michael A. Caplinger, Brandi L. Carrier,
Nataly Chen, Amy Chen, Matthew J. Clark, Samuel Clegg, Pamela G. Conrad, Moogega Cooper, Kristine N. Davis, Bethany
Ehlmann, Linda Facto, Marc D. Fries, Dan H. Garrison, Denine Gasway, F. Tony Ghaemi, Trevor G. Graff, Kevin P. Hand,
Cathleen Harris, Jeffrey D. Hein, Nicholas Heinz, Harrison Herzog, Eric Hochberg, Andrew Houck, William F. Hug, Elsa H.
Jensen, Linda C. Kah, John Kennedy, Robert Krylo, Johnathan Lam, Mark Lindeman, Justin McGlown, John Michel, Ed Miller,
Zachary Mills, Michelle E. Minitti, Fai Mok, James Moore, Kenneth H. Nealson, Anthony Nelson, Raymond Newell, Brian E.
Nixon, Daniel A. Nordman, Danielle Nuding, Sonny Orellana, Michael Pauken, Glen Peterson, Randy Pollock, Heather Quinn,
Claire Quinto, Michael A. Ravine, Ray D. Reid, Joe Riendeau, Amy J. Ross, Joshua Sackos, Jacob A. Schaffner, Mark Schwochert,
Molly O Shelton, Rufus Simon, Caroline L. Smith, Pablo Sobron, Kimberly Steadman, Andrew Steele, Dave Thiessen, Vinh
D. Tran, Tony Tsai, Michael Tuite, Eric Tung, Rami Wehbe, Rachel Weinberg, Ryan H. Weiner, Roger C. Wiens, Kenneth
Williford, Chris Wollonciej, Yen-Hung Wu, R. Aileen Yingst, Jason Zan. 2021. Perseverance’s Scanning Habitable Environments
with Raman and Luminescence for Organics and Chemicals (SHERLOC) Investigation. Space Science Reviews 217:4. . [Crossref]